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/rtc.h>
  55 #include <asm/uv/uv.h>
  56
  57 #define EFI_DEBUG
  58
  59 struct efi_memory_map memmap;
  60
  61 static struct efi efi_phys __initdata;
  62 static efi_system_table_t efi_systab __initdata;
  63
  64 static efi_config_table_type_t arch_tables[] __initdata = {
  65 #ifdef CONFIG_X86_UV
  66         {UV_SYSTEM_TABLE_GUID, "UVsystab", &efi.uv_systab},
  67 #endif
  68         {NULL_GUID, NULL, NULL},
  69 };
  70
  71 u64 efi_setup;          /* efi setup_data physical address */
  72
  73 static int add_efi_memmap __initdata;
  74 static int __init setup_add_efi_memmap(char *arg)
  75 {
  76         add_efi_memmap = 1;
  77         return 0;
  78 }
  79 early_param("add_efi_memmap", setup_add_efi_memmap);
  80
  81 static efi_status_t __init phys_efi_set_virtual_address_map(
  82         unsigned long memory_map_size,
  83         unsigned long descriptor_size,
  84         u32 descriptor_version,
  85         efi_memory_desc_t *virtual_map)
  86 {
  87         efi_status_t status;
  88         unsigned long flags;
  89         pgd_t *save_pgd;
  90
  91         save_pgd = efi_call_phys_prolog();
  92
  93         /* Disable interrupts around EFI calls: */
  94         local_irq_save(flags);
  95         status = efi_call_phys(efi_phys.set_virtual_address_map,
  96                                memory_map_size, descriptor_size,
  97                                descriptor_version, virtual_map);
  98         local_irq_restore(flags);
  99
 100         efi_call_phys_epilog(save_pgd);
 101
 102         return status;
 103 }
 104
 105 void efi_get_time(struct timespec *now)
 106 {
 107         efi_status_t status;
 108         efi_time_t eft;
 109         efi_time_cap_t cap;
 110
 111         status = efi.get_time(&eft, &cap);
 112         if (status != EFI_SUCCESS)
 113                 pr_err("Oops: efitime: can't read time!\n");
 114
 115         now->tv_sec = mktime(eft.year, eft.month, eft.day, eft.hour,
 116                              eft.minute, eft.second);
 117         now->tv_nsec = 0;
 118 }
 119
 120 void __init efi_find_mirror(void)
 121 {
 122         void *p;
 123         u64 mirror_size = 0, total_size = 0;
 124
 125         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 126                 efi_memory_desc_t *md = p;
 127                 unsigned long long start = md->phys_addr;
 128                 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
 129
 130                 total_size += size;
 131                 if (md->attribute & EFI_MEMORY_MORE_RELIABLE) {
 132                         memblock_mark_mirror(start, size);
 133                         mirror_size += size;
 134                 }
 135         }
 136         if (mirror_size)
 137                 pr_info("Memory: %lldM/%lldM mirrored memory\n",
 138                         mirror_size>>20, total_size>>20);
 139 }
 140
 141 /*
 142  * Tell the kernel about the EFI memory map.  This might include
 143  * more than the max 128 entries that can fit in the e820 legacy
 144  * (zeropage) memory map.
 145  */
 146
 147 static void __init do_add_efi_memmap(void)
 148 {
 149         void *p;
 150
 151         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 152                 efi_memory_desc_t *md = p;
 153                 unsigned long long start = md->phys_addr;
 154                 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
 155                 int e820_type;
 156
 157                 switch (md->type) {
 158                 case EFI_LOADER_CODE:
 159                 case EFI_LOADER_DATA:
 160                 case EFI_BOOT_SERVICES_CODE:
 161                 case EFI_BOOT_SERVICES_DATA:
 162                 case EFI_CONVENTIONAL_MEMORY:
 163                         if (md->attribute & EFI_MEMORY_WB)
 164                                 e820_type = E820_RAM;
 165                         else
 166                                 e820_type = E820_RESERVED;
 167                         break;
 168                 case EFI_ACPI_RECLAIM_MEMORY:
 169                         e820_type = E820_ACPI;
 170                         break;
 171                 case EFI_ACPI_MEMORY_NVS:
 172                         e820_type = E820_NVS;
 173                         break;
 174                 case EFI_UNUSABLE_MEMORY:
 175                         e820_type = E820_UNUSABLE;
 176                         break;
 177                 case EFI_PERSISTENT_MEMORY:
 178                         e820_type = E820_PMEM;
 179                         break;
 180                 default:
 181                         /*
 182                          * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
 183                          * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
 184                          * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
 185                          */
 186                         e820_type = E820_RESERVED;
 187                         break;
 188                 }
 189                 e820_add_region(start, size, e820_type);
 190         }
 191         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 192 }
 193
 194 int __init efi_memblock_x86_reserve_range(void)
 195 {
 196         struct efi_info *e = &boot_params.efi_info;
 197         phys_addr_t pmap;
 198
 199         if (efi_enabled(EFI_PARAVIRT))
 200                 return 0;
 201
 202 #ifdef CONFIG_X86_32
 203         /* Can't handle data above 4GB at this time */
 204         if (e->efi_memmap_hi) {
 205                 pr_err("Memory map is above 4GB, disabling EFI.\n");
 206                 return -EINVAL;
 207         }
 208         pmap =  e->efi_memmap;
 209 #else
 210         pmap = (e->efi_memmap | ((__u64)e->efi_memmap_hi << 32));
 211 #endif
 212         memmap.phys_map         = pmap;
 213         memmap.nr_map           = e->efi_memmap_size /
 214                                   e->efi_memdesc_size;
 215         memmap.desc_size        = e->efi_memdesc_size;
 216         memmap.desc_version     = e->efi_memdesc_version;
 217
 218         memblock_reserve(pmap, memmap.nr_map * memmap.desc_size);
 219
 220         efi.memmap = &memmap;
 221
 222         return 0;
 223 }
 224
 225 void __init efi_print_memmap(void)
 226 {
 227 #ifdef EFI_DEBUG
 228         efi_memory_desc_t *md;
 229         void *p;
 230         int i;
 231
 232         for (p = memmap.map, i = 0;
 233              p < memmap.map_end;
 234              p += memmap.desc_size, i++) {
 235                 char buf[64];
 236
 237                 md = p;
 238                 pr_info("mem%02u: %s range=[0x%016llx-0x%016llx) (%lluMB)\n",
 239                         i, efi_md_typeattr_format(buf, sizeof(buf), md),
 240                         md->phys_addr,
 241                         md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT),
 242                         (md->num_pages >> (20 - EFI_PAGE_SHIFT)));
 243         }
 244 #endif  /*  EFI_DEBUG  */
 245 }
 246
 247 void __init efi_unmap_memmap(void)
 248 {
 249         clear_bit(EFI_MEMMAP, &efi.flags);
 250         if (memmap.map) {
 251                 early_memunmap(memmap.map, memmap.nr_map * memmap.desc_size);
 252                 memmap.map = NULL;
 253         }
 254 }
 255
 256 static int __init efi_systab_init(void *phys)
 257 {
 258         if (efi_enabled(EFI_64BIT)) {
 259                 efi_system_table_64_t *systab64;
 260                 struct efi_setup_data *data = NULL;
 261                 u64 tmp = 0;
 262
 263                 if (efi_setup) {
 264                         data = early_memremap(efi_setup, sizeof(*data));
 265                         if (!data)
 266                                 return -ENOMEM;
 267                 }
 268                 systab64 = early_memremap((unsigned long)phys,
 269                                          sizeof(*systab64));
 270                 if (systab64 == NULL) {
 271                         pr_err("Couldn't map the system table!\n");
 272                         if (data)
 273                                 early_memunmap(data, sizeof(*data));
 274                         return -ENOMEM;
 275                 }
 276
 277                 efi_systab.hdr = systab64->hdr;
 278                 efi_systab.fw_vendor = data ? (unsigned long)data->fw_vendor :
 279                                               systab64->fw_vendor;
 280                 tmp |= data ? data->fw_vendor : systab64->fw_vendor;
 281                 efi_systab.fw_revision = systab64->fw_revision;
 282                 efi_systab.con_in_handle = systab64->con_in_handle;
 283                 tmp |= systab64->con_in_handle;
 284                 efi_systab.con_in = systab64->con_in;
 285                 tmp |= systab64->con_in;
 286                 efi_systab.con_out_handle = systab64->con_out_handle;
 287                 tmp |= systab64->con_out_handle;
 288                 efi_systab.con_out = systab64->con_out;
 289                 tmp |= systab64->con_out;
 290                 efi_systab.stderr_handle = systab64->stderr_handle;
 291                 tmp |= systab64->stderr_handle;
 292                 efi_systab.stderr = systab64->stderr;
 293                 tmp |= systab64->stderr;
 294                 efi_systab.runtime = data ?
 295                                      (void *)(unsigned long)data->runtime :
 296                                      (void *)(unsigned long)systab64->runtime;
 297                 tmp |= data ? data->runtime : systab64->runtime;
 298                 efi_systab.boottime = (void *)(unsigned long)systab64->boottime;
 299                 tmp |= systab64->boottime;
 300                 efi_systab.nr_tables = systab64->nr_tables;
 301                 efi_systab.tables = data ? (unsigned long)data->tables :
 302                                            systab64->tables;
 303                 tmp |= data ? data->tables : systab64->tables;
 304
 305                 early_memunmap(systab64, sizeof(*systab64));
 306                 if (data)
 307                         early_memunmap(data, sizeof(*data));
 308 #ifdef CONFIG_X86_32
 309                 if (tmp >> 32) {
 310                         pr_err("EFI data located above 4GB, disabling EFI.\n");
 311                         return -EINVAL;
 312                 }
 313 #endif
 314         } else {
 315                 efi_system_table_32_t *systab32;
 316
 317                 systab32 = early_memremap((unsigned long)phys,
 318                                          sizeof(*systab32));
 319                 if (systab32 == NULL) {
 320                         pr_err("Couldn't map the system table!\n");
 321                         return -ENOMEM;
 322                 }
 323
 324                 efi_systab.hdr = systab32->hdr;
 325                 efi_systab.fw_vendor = systab32->fw_vendor;
 326                 efi_systab.fw_revision = systab32->fw_revision;
 327                 efi_systab.con_in_handle = systab32->con_in_handle;
 328                 efi_systab.con_in = systab32->con_in;
 329                 efi_systab.con_out_handle = systab32->con_out_handle;
 330                 efi_systab.con_out = systab32->con_out;
 331                 efi_systab.stderr_handle = systab32->stderr_handle;
 332                 efi_systab.stderr = systab32->stderr;
 333                 efi_systab.runtime = (void *)(unsigned long)systab32->runtime;
 334                 efi_systab.boottime = (void *)(unsigned long)systab32->boottime;
 335                 efi_systab.nr_tables = systab32->nr_tables;
 336                 efi_systab.tables = systab32->tables;
 337
 338                 early_memunmap(systab32, sizeof(*systab32));
 339         }
 340
 341         efi.systab = &efi_systab;
 342
 343         /*
 344          * Verify the EFI Table
 345          */
 346         if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE) {
 347                 pr_err("System table signature incorrect!\n");
 348                 return -EINVAL;
 349         }
 350         if ((efi.systab->hdr.revision >> 16) == 0)
 351                 pr_err("Warning: System table version %d.%02d, expected 1.00 or greater!\n",
 352                        efi.systab->hdr.revision >> 16,
 353                        efi.systab->hdr.revision & 0xffff);
 354
 355         set_bit(EFI_SYSTEM_TABLES, &efi.flags);
 356
 357         return 0;
 358 }
 359
 360 static int __init efi_runtime_init32(void)
 361 {
 362         efi_runtime_services_32_t *runtime;
 363
 364         runtime = early_memremap((unsigned long)efi.systab->runtime,
 365                         sizeof(efi_runtime_services_32_t));
 366         if (!runtime) {
 367                 pr_err("Could not map the runtime service table!\n");
 368                 return -ENOMEM;
 369         }
 370
 371         /*
 372          * We will only need *early* access to the SetVirtualAddressMap
 373          * EFI runtime service. All other runtime services will be called
 374          * via the virtual mapping.
 375          */
 376         efi_phys.set_virtual_address_map =
 377                         (efi_set_virtual_address_map_t *)
 378                         (unsigned long)runtime->set_virtual_address_map;
 379         early_memunmap(runtime, sizeof(efi_runtime_services_32_t));
 380
 381         return 0;
 382 }
 383
 384 static int __init efi_runtime_init64(void)
 385 {
 386         efi_runtime_services_64_t *runtime;
 387
 388         runtime = early_memremap((unsigned long)efi.systab->runtime,
 389                         sizeof(efi_runtime_services_64_t));
 390         if (!runtime) {
 391                 pr_err("Could not map the runtime service table!\n");
 392                 return -ENOMEM;
 393         }
 394
 395         /*
 396          * We will only need *early* access to the SetVirtualAddressMap
 397          * EFI runtime service. All other runtime services will be called
 398          * via the virtual mapping.
 399          */
 400         efi_phys.set_virtual_address_map =
 401                         (efi_set_virtual_address_map_t *)
 402                         (unsigned long)runtime->set_virtual_address_map;
 403         early_memunmap(runtime, sizeof(efi_runtime_services_64_t));
 404
 405         return 0;
 406 }
 407
 408 static int __init efi_runtime_init(void)
 409 {
 410         int rv;
 411
 412         /*
 413          * Check out the runtime services table. We need to map
 414          * the runtime services table so that we can grab the physical
 415          * address of several of the EFI runtime functions, needed to
 416          * set the firmware into virtual mode.
 417          *
 418          * When EFI_PARAVIRT is in force then we could not map runtime
 419          * service memory region because we do not have direct access to it.
 420          * However, runtime services are available through proxy functions
 421          * (e.g. in case of Xen dom0 EFI implementation they call special
 422          * hypercall which executes relevant EFI functions) and that is why
 423          * they are always enabled.
 424          */
 425
 426         if (!efi_enabled(EFI_PARAVIRT)) {
 427                 if (efi_enabled(EFI_64BIT))
 428                         rv = efi_runtime_init64();
 429                 else
 430                         rv = efi_runtime_init32();
 431
 432                 if (rv)
 433                         return rv;
 434         }
 435
 436         set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
 437
 438         return 0;
 439 }
 440
 441 static int __init efi_memmap_init(void)
 442 {
 443         if (efi_enabled(EFI_PARAVIRT))
 444                 return 0;
 445
 446         /* Map the EFI memory map */
 447         memmap.map = early_memremap((unsigned long)memmap.phys_map,
 448                                    memmap.nr_map * memmap.desc_size);
 449         if (memmap.map == NULL) {
 450                 pr_err("Could not map the memory map!\n");
 451                 return -ENOMEM;
 452         }
 453         memmap.map_end = memmap.map + (memmap.nr_map * memmap.desc_size);
 454
 455         if (add_efi_memmap)
 456                 do_add_efi_memmap();
 457
 458         set_bit(EFI_MEMMAP, &efi.flags);
 459
 460         return 0;
 461 }
 462
 463 void __init efi_init(void)
 464 {
 465         efi_char16_t *c16;
 466         char vendor[100] = "unknown";
 467         int i = 0;
 468
 469 #ifdef CONFIG_X86_32
 470         if (boot_params.efi_info.efi_systab_hi ||
 471             boot_params.efi_info.efi_memmap_hi) {
 472                 pr_info("Table located above 4GB, disabling EFI.\n");
 473                 return;
 474         }
 475         efi_phys.systab = (efi_system_table_t *)boot_params.efi_info.efi_systab;
 476 #else
 477         efi_phys.systab = (efi_system_table_t *)
 478                           (boot_params.efi_info.efi_systab |
 479                           ((__u64)boot_params.efi_info.efi_systab_hi<<32));
 480 #endif
 481
 482         if (efi_systab_init(efi_phys.systab))
 483                 return;
 484
 485         efi.config_table = (unsigned long)efi.systab->tables;
 486         efi.fw_vendor    = (unsigned long)efi.systab->fw_vendor;
 487         efi.runtime      = (unsigned long)efi.systab->runtime;
 488
 489         /*
 490          * Show what we know for posterity
 491          */
 492         c16 = early_memremap_ro(efi.systab->fw_vendor,
 493                                 sizeof(vendor) * sizeof(efi_char16_t));
 494         if (c16) {
 495                 for (i = 0; i < sizeof(vendor) - 1 && c16[i]; ++i)
 496                         vendor[i] = c16[i];
 497                 vendor[i] = '\0';
 498                 early_memunmap(c16, sizeof(vendor) * sizeof(efi_char16_t));
 499         } else {
 500                 pr_err("Could not map the firmware vendor!\n");
 501         }
 502
 503         pr_info("EFI v%u.%.02u by %s\n",
 504                 efi.systab->hdr.revision >> 16,
 505                 efi.systab->hdr.revision & 0xffff, vendor);
 506
 507         if (efi_reuse_config(efi.systab->tables, efi.systab->nr_tables))
 508                 return;
 509
 510         if (efi_config_init(arch_tables))
 511                 return;
 512
 513         /*
 514          * Note: We currently don't support runtime services on an EFI
 515          * that doesn't match the kernel 32/64-bit mode.
 516          */
 517
 518         if (!efi_runtime_supported())
 519                 pr_info("No EFI runtime due to 32/64-bit mismatch with kernel\n");
 520         else {
 521                 if (efi_runtime_disabled() || efi_runtime_init())
 522                         return;
 523         }
 524         if (efi_memmap_init())
 525                 return;
 526
 527         if (efi_enabled(EFI_DBG))
 528                 efi_print_memmap();
 529
 530         efi_esrt_init();
 531 }
 532
 533 void __init efi_late_init(void)
 534 {
 535         efi_bgrt_init();
 536 }
 537
 538 void __init efi_set_executable(efi_memory_desc_t *md, bool executable)
 539 {
 540         u64 addr, npages;
 541
 542         addr = md->virt_addr;
 543         npages = md->num_pages;
 544
 545         memrange_efi_to_native(&addr, &npages);
 546
 547         if (executable)
 548                 set_memory_x(addr, npages);
 549         else
 550                 set_memory_nx(addr, npages);
 551 }
 552
 553 void __init runtime_code_page_mkexec(void)
 554 {
 555         efi_memory_desc_t *md;
 556         void *p;
 557
 558         /* Make EFI runtime service code area executable */
 559         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 560                 md = p;
 561
 562                 if (md->type != EFI_RUNTIME_SERVICES_CODE)
 563                         continue;
 564
 565                 efi_set_executable(md, true);
 566         }
 567 }
 568
 569 void __init efi_memory_uc(u64 addr, unsigned long size)
 570 {
 571         unsigned long page_shift = 1UL << EFI_PAGE_SHIFT;
 572         u64 npages;
 573
 574         npages = round_up(size, page_shift) / page_shift;
 575         memrange_efi_to_native(&addr, &npages);
 576         set_memory_uc(addr, npages);
 577 }
 578
 579 void __init old_map_region(efi_memory_desc_t *md)
 580 {
 581         u64 start_pfn, end_pfn, end;
 582         unsigned long size;
 583         void *va;
 584
 585         start_pfn = PFN_DOWN(md->phys_addr);
 586         size      = md->num_pages << PAGE_SHIFT;
 587         end       = md->phys_addr + size;
 588         end_pfn   = PFN_UP(end);
 589
 590         if (pfn_range_is_mapped(start_pfn, end_pfn)) {
 591                 va = __va(md->phys_addr);
 592
 593                 if (!(md->attribute & EFI_MEMORY_WB))
 594                         efi_memory_uc((u64)(unsigned long)va, size);
 595         } else
 596                 va = efi_ioremap(md->phys_addr, size,
 597                                  md->type, md->attribute);
 598
 599         md->virt_addr = (u64) (unsigned long) va;
 600         if (!va)
 601                 pr_err("ioremap of 0x%llX failed!\n",
 602                        (unsigned long long)md->phys_addr);
 603 }
 604
 605 /* Merge contiguous regions of the same type and attribute */
 606 static void __init efi_merge_regions(void)
 607 {
 608         void *p;
 609         efi_memory_desc_t *md, *prev_md = NULL;
 610
 611         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 612                 u64 prev_size;
 613                 md = p;
 614
 615                 if (!prev_md) {
 616                         prev_md = md;
 617                         continue;
 618                 }
 619
 620                 if (prev_md->type != md->type ||
 621                     prev_md->attribute != md->attribute) {
 622                         prev_md = md;
 623                         continue;
 624                 }
 625
 626                 prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
 627
 628                 if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
 629                         prev_md->num_pages += md->num_pages;
 630                         md->type = EFI_RESERVED_TYPE;
 631                         md->attribute = 0;
 632                         continue;
 633                 }
 634                 prev_md = md;
 635         }
 636 }
 637
 638 static void __init get_systab_virt_addr(efi_memory_desc_t *md)
 639 {
 640         unsigned long size;
 641         u64 end, systab;
 642
 643         size = md->num_pages << EFI_PAGE_SHIFT;
 644         end = md->phys_addr + size;
 645         systab = (u64)(unsigned long)efi_phys.systab;
 646         if (md->phys_addr <= systab && systab < end) {
 647                 systab += md->virt_addr - md->phys_addr;
 648                 efi.systab = (efi_system_table_t *)(unsigned long)systab;
 649         }
 650 }
 651
 652 static void __init save_runtime_map(void)
 653 {
 654 #ifdef CONFIG_KEXEC_CORE
 655         efi_memory_desc_t *md;
 656         void *tmp, *p, *q = NULL;
 657         int count = 0;
 658
 659         if (efi_enabled(EFI_OLD_MEMMAP))
 660                 return;
 661
 662         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 663                 md = p;
 664
 665                 if (!(md->attribute & EFI_MEMORY_RUNTIME) ||
 666                     (md->type == EFI_BOOT_SERVICES_CODE) ||
 667                     (md->type == EFI_BOOT_SERVICES_DATA))
 668                         continue;
 669                 tmp = krealloc(q, (count + 1) * memmap.desc_size, GFP_KERNEL);
 670                 if (!tmp)
 671                         goto out;
 672                 q = tmp;
 673
 674                 memcpy(q + count * memmap.desc_size, md, memmap.desc_size);
 675                 count++;
 676         }
 677
 678         efi_runtime_map_setup(q, count, memmap.desc_size);
 679         return;
 680
 681 out:
 682         kfree(q);
 683         pr_err("Error saving runtime map, efi runtime on kexec non-functional!!\n");
 684 #endif
 685 }
 686
 687 static void *realloc_pages(void *old_memmap, int old_shift)
 688 {
 689         void *ret;
 690
 691         ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1);
 692         if (!ret)
 693                 goto out;
 694
 695         /*
 696          * A first-time allocation doesn't have anything to copy.
 697          */
 698         if (!old_memmap)
 699                 return ret;
 700
 701         memcpy(ret, old_memmap, PAGE_SIZE << old_shift);
 702
 703 out:
 704         free_pages((unsigned long)old_memmap, old_shift);
 705         return ret;
 706 }
 707
 708 /*
 709  * Iterate the EFI memory map in reverse order because the regions
 710  * will be mapped top-down. The end result is the same as if we had
 711  * mapped things forward, but doesn't require us to change the
 712  * existing implementation of efi_map_region().
 713  */
 714 static inline void *efi_map_next_entry_reverse(void *entry)
 715 {
 716         /* Initial call */
 717         if (!entry)
 718                 return memmap.map_end - memmap.desc_size;
 719
 720         entry -= memmap.desc_size;
 721         if (entry < memmap.map)
 722                 return NULL;
 723
 724         return entry;
 725 }
 726
 727 /*
 728  * efi_map_next_entry - Return the next EFI memory map descriptor
 729  * @entry: Previous EFI memory map descriptor
 730  *
 731  * This is a helper function to iterate over the EFI memory map, which
 732  * we do in different orders depending on the current configuration.
 733  *
 734  * To begin traversing the memory map @entry must be %NULL.
 735  *
 736  * Returns %NULL when we reach the end of the memory map.
 737  */
 738 static void *efi_map_next_entry(void *entry)
 739 {
 740         if (!efi_enabled(EFI_OLD_MEMMAP) && efi_enabled(EFI_64BIT)) {
 741                 /*
 742                  * Starting in UEFI v2.5 the EFI_PROPERTIES_TABLE
 743                  * config table feature requires us to map all entries
 744                  * in the same order as they appear in the EFI memory
 745                  * map. That is to say, entry N must have a lower
 746                  * virtual address than entry N+1. This is because the
 747                  * firmware toolchain leaves relative references in
 748                  * the code/data sections, which are split and become
 749                  * separate EFI memory regions. Mapping things
 750                  * out-of-order leads to the firmware accessing
 751                  * unmapped addresses.
 752                  *
 753                  * Since we need to map things this way whether or not
 754                  * the kernel actually makes use of
 755                  * EFI_PROPERTIES_TABLE, let's just switch to this
 756                  * scheme by default for 64-bit.
 757                  */
 758                 return efi_map_next_entry_reverse(entry);
 759         }
 760
 761         /* Initial call */
 762         if (!entry)
 763                 return memmap.map;
 764
 765         entry += memmap.desc_size;
 766         if (entry >= memmap.map_end)
 767                 return NULL;
 768
 769         return entry;
 770 }
 771
 772 /*
 773  * Map the efi memory ranges of the runtime services and update new_mmap with
 774  * virtual addresses.
 775  */
 776 static void * __init efi_map_regions(int *count, int *pg_shift)
 777 {
 778         void *p, *new_memmap = NULL;
 779         unsigned long left = 0;
 780         efi_memory_desc_t *md;
 781
 782         p = NULL;
 783         while ((p = efi_map_next_entry(p))) {
 784                 md = p;
 785                 if (!(md->attribute & EFI_MEMORY_RUNTIME)) {
 786 #ifdef CONFIG_X86_64
 787                         if (md->type != EFI_BOOT_SERVICES_CODE &&
 788                             md->type != EFI_BOOT_SERVICES_DATA)
 789 #endif
 790                                 continue;
 791                 }
 792
 793                 efi_map_region(md);
 794                 get_systab_virt_addr(md);
 795
 796                 if (left < memmap.desc_size) {
 797                         new_memmap = realloc_pages(new_memmap, *pg_shift);
 798                         if (!new_memmap)
 799                                 return NULL;
 800
 801                         left += PAGE_SIZE << *pg_shift;
 802                         (*pg_shift)++;
 803                 }
 804
 805                 memcpy(new_memmap + (*count * memmap.desc_size), md,
 806                        memmap.desc_size);
 807
 808                 left -= memmap.desc_size;
 809                 (*count)++;
 810         }
 811
 812         return new_memmap;
 813 }
 814
 815 static void __init kexec_enter_virtual_mode(void)
 816 {
 817 #ifdef CONFIG_KEXEC_CORE
 818         efi_memory_desc_t *md;
 819         void *p;
 820
 821         efi.systab = NULL;
 822
 823         /*
 824          * We don't do virtual mode, since we don't do runtime services, on
 825          * non-native EFI
 826          */
 827         if (!efi_is_native()) {
 828                 efi_unmap_memmap();
 829                 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
 830                 return;
 831         }
 832
 833         /*
 834         * Map efi regions which were passed via setup_data. The virt_addr is a
 835         * fixed addr which was used in first kernel of a kexec boot.
 836         */
 837         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 838                 md = p;
 839                 efi_map_region_fixed(md); /* FIXME: add error handling */
 840                 get_systab_virt_addr(md);
 841         }
 842
 843         save_runtime_map();
 844
 845         BUG_ON(!efi.systab);
 846
 847         efi_sync_low_kernel_mappings();
 848
 849         /*
 850          * Now that EFI is in virtual mode, update the function
 851          * pointers in the runtime service table to the new virtual addresses.
 852          *
 853          * Call EFI services through wrapper functions.
 854          */
 855         efi.runtime_version = efi_systab.hdr.revision;
 856
 857         efi_native_runtime_setup();
 858
 859         efi.set_virtual_address_map = NULL;
 860
 861         if (efi_enabled(EFI_OLD_MEMMAP) && (__supported_pte_mask & _PAGE_NX))
 862                 runtime_code_page_mkexec();
 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  * ->trampoline_pgd page table using a top-down VA allocation scheme.
 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 with efi=old_map on its command line. Same old
 875  * method enabled the runtime services to be called without having to
 876  * thunk back into 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 PGDs of the
 881  * kernel page table into ->trampoline_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
 894         efi.systab = NULL;
 895
 896         efi_merge_regions();
 897         new_memmap = efi_map_regions(&count, &pg_shift);
 898         if (!new_memmap) {
 899                 pr_err("Error reallocating memory, EFI runtime non-functional!\n");
 900                 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
 901                 return;
 902         }
 903
 904         save_runtime_map();
 905
 906         BUG_ON(!efi.systab);
 907
 908         if (efi_setup_page_tables(__pa(new_memmap), 1 << pg_shift)) {
 909                 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
 910                 return;
 911         }
 912
 913         efi_sync_low_kernel_mappings();
 914         efi_dump_pagetable();
 915
 916         if (efi_is_native()) {
 917                 status = phys_efi_set_virtual_address_map(
 918                                 memmap.desc_size * count,
 919                                 memmap.desc_size,
 920                                 memmap.desc_version,
 921                                 (efi_memory_desc_t *)__pa(new_memmap));
 922         } else {
 923                 status = efi_thunk_set_virtual_address_map(
 924                                 efi_phys.set_virtual_address_map,
 925                                 memmap.desc_size * count,
 926                                 memmap.desc_size,
 927                                 memmap.desc_version,
 928                                 (efi_memory_desc_t *)__pa(new_memmap));
 929         }
 930
 931         if (status != EFI_SUCCESS) {
 932                 pr_alert("Unable to switch EFI into virtual mode (status=%lx)!\n",
 933                          status);
 934                 panic("EFI call to SetVirtualAddressMap() failed!");
 935         }
 936
 937         /*
 938          * Now that EFI is in virtual mode, update the function
 939          * pointers in the runtime service table to the new virtual addresses.
 940          *
 941          * Call EFI services through wrapper functions.
 942          */
 943         efi.runtime_version = efi_systab.hdr.revision;
 944
 945         if (efi_is_native())
 946                 efi_native_runtime_setup();
 947         else
 948                 efi_thunk_runtime_setup();
 949
 950         efi.set_virtual_address_map = NULL;
 951
 952         efi_runtime_mkexec();
 953
 954         /*
 955          * We mapped the descriptor array into the EFI pagetable above but we're
 956          * not unmapping it here. Here's why:
 957          *
 958          * We're copying select PGDs from the kernel page table to the EFI page
 959          * table and when we do so and make changes to those PGDs like unmapping
 960          * stuff from them, those changes appear in the kernel page table and we
 961          * go boom.
 962          *
 963          * From setup_real_mode():
 964          *
 965          * ...
 966          * trampoline_pgd[0] = init_level4_pgt[pgd_index(__PAGE_OFFSET)].pgd;
 967          *
 968          * In this particular case, our allocation is in PGD 0 of the EFI page
 969          * table but we've copied that PGD from PGD[272] of the EFI page table:
 970          *
 971          *      pgd_index(__PAGE_OFFSET = 0xffff880000000000) = 272
 972          *
 973          * where the direct memory mapping in kernel space is.
 974          *
 975          * new_memmap's VA comes from that direct mapping and thus clearing it,
 976          * it would get cleared in the kernel page table too.
 977          *
 978          * efi_cleanup_page_tables(__pa(new_memmap), 1 << pg_shift);
 979          */
 980         free_pages((unsigned long)new_memmap, pg_shift);
 981
 982         /* clean DUMMY object */
 983         efi_delete_dummy_variable();
 984 }
 985
 986 void __init efi_enter_virtual_mode(void)
 987 {
 988         if (efi_enabled(EFI_PARAVIRT))
 989                 return;
 990
 991         if (efi_setup)
 992                 kexec_enter_virtual_mode();
 993         else
 994                 __efi_enter_virtual_mode();
 995 }
 996
 997 /*
 998  * Convenience functions to obtain memory types and attributes
 999  */
1000 u32 efi_mem_type(unsigned long phys_addr)
1001 {
1002         efi_memory_desc_t *md;
1003         void *p;
1004
1005         if (!efi_enabled(EFI_MEMMAP))
1006                 return 0;
1007
1008         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
1009                 md = p;
1010                 if ((md->phys_addr <= phys_addr) &&
1011                     (phys_addr < (md->phys_addr +
1012                                   (md->num_pages << EFI_PAGE_SHIFT))))
1013                         return md->type;
1014         }
1015         return 0;
1016 }
1017
1018 static int __init arch_parse_efi_cmdline(char *str)
1019 {
1020         if (!str) {
1021                 pr_warn("need at least one option\n");
1022                 return -EINVAL;
1023         }
1024
1025         if (parse_option_str(str, "old_map"))
1026                 set_bit(EFI_OLD_MEMMAP, &efi.flags);
1027
1028         return 0;
1029 }
1030 early_param("efi", arch_parse_efi_cmdline);