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
  56 #define EFI_DEBUG
  57
  58 #define EFI_MIN_RESERVE 5120
  59
  60 #define EFI_DUMMY_GUID \
  61         EFI_GUID(0x4424ac57, 0xbe4b, 0x47dd, 0x9e, 0x97, 0xed, 0x50, 0xf0, 0x9f, 0x92, 0xa9)
  62
  63 static efi_char16_t efi_dummy_name[6] = { 'D', 'U', 'M', 'M', 'Y', 0 };
  64
  65 struct efi_memory_map memmap;
  66
  67 static struct efi efi_phys __initdata;
  68 static efi_system_table_t efi_systab __initdata;
  69
  70 unsigned long x86_efi_facility;
  71
  72 static __initdata efi_config_table_type_t arch_tables[] = {
  73 #ifdef CONFIG_X86_UV
  74         {UV_SYSTEM_TABLE_GUID, "UVsystab", &efi.uv_systab},
  75 #endif
  76         {NULL_GUID, NULL, NULL},
  77 };
  78
  79 u64 efi_setup;          /* efi setup_data physical address */
  80
  81 /*
  82  * Returns 1 if 'facility' is enabled, 0 otherwise.
  83  */
  84 int efi_enabled(int facility)
  85 {
  86         return test_bit(facility, &x86_efi_facility) != 0;
  87 }
  88 EXPORT_SYMBOL(efi_enabled);
  89
  90 static bool __initdata disable_runtime = false;
  91 static int __init setup_noefi(char *arg)
  92 {
  93         disable_runtime = true;
  94         return 0;
  95 }
  96 early_param("noefi", setup_noefi);
  97
  98 int add_efi_memmap;
  99 EXPORT_SYMBOL(add_efi_memmap);
 100
 101 static int __init setup_add_efi_memmap(char *arg)
 102 {
 103         add_efi_memmap = 1;
 104         return 0;
 105 }
 106 early_param("add_efi_memmap", setup_add_efi_memmap);
 107
 108 static bool efi_no_storage_paranoia;
 109
 110 static int __init setup_storage_paranoia(char *arg)
 111 {
 112         efi_no_storage_paranoia = true;
 113         return 0;
 114 }
 115 early_param("efi_no_storage_paranoia", setup_storage_paranoia);
 116
 117 static efi_status_t virt_efi_get_time(efi_time_t *tm, efi_time_cap_t *tc)
 118 {
 119         unsigned long flags;
 120         efi_status_t status;
 121
 122         spin_lock_irqsave(&rtc_lock, flags);
 123         status = efi_call_virt2(get_time, tm, tc);
 124         spin_unlock_irqrestore(&rtc_lock, flags);
 125         return status;
 126 }
 127
 128 static efi_status_t virt_efi_set_time(efi_time_t *tm)
 129 {
 130         unsigned long flags;
 131         efi_status_t status;
 132
 133         spin_lock_irqsave(&rtc_lock, flags);
 134         status = efi_call_virt1(set_time, tm);
 135         spin_unlock_irqrestore(&rtc_lock, flags);
 136         return status;
 137 }
 138
 139 static efi_status_t virt_efi_get_wakeup_time(efi_bool_t *enabled,
 140                                              efi_bool_t *pending,
 141                                              efi_time_t *tm)
 142 {
 143         unsigned long flags;
 144         efi_status_t status;
 145
 146         spin_lock_irqsave(&rtc_lock, flags);
 147         status = efi_call_virt3(get_wakeup_time,
 148                                 enabled, pending, tm);
 149         spin_unlock_irqrestore(&rtc_lock, flags);
 150         return status;
 151 }
 152
 153 static efi_status_t virt_efi_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm)
 154 {
 155         unsigned long flags;
 156         efi_status_t status;
 157
 158         spin_lock_irqsave(&rtc_lock, flags);
 159         status = efi_call_virt2(set_wakeup_time,
 160                                 enabled, tm);
 161         spin_unlock_irqrestore(&rtc_lock, flags);
 162         return status;
 163 }
 164
 165 static efi_status_t virt_efi_get_variable(efi_char16_t *name,
 166                                           efi_guid_t *vendor,
 167                                           u32 *attr,
 168                                           unsigned long *data_size,
 169                                           void *data)
 170 {
 171         return efi_call_virt5(get_variable,
 172                               name, vendor, attr,
 173                               data_size, data);
 174 }
 175
 176 static efi_status_t virt_efi_get_next_variable(unsigned long *name_size,
 177                                                efi_char16_t *name,
 178                                                efi_guid_t *vendor)
 179 {
 180         return efi_call_virt3(get_next_variable,
 181                               name_size, name, vendor);
 182 }
 183
 184 static efi_status_t virt_efi_set_variable(efi_char16_t *name,
 185                                           efi_guid_t *vendor,
 186                                           u32 attr,
 187                                           unsigned long data_size,
 188                                           void *data)
 189 {
 190         return efi_call_virt5(set_variable,
 191                               name, vendor, attr,
 192                               data_size, data);
 193 }
 194
 195 static efi_status_t virt_efi_query_variable_info(u32 attr,
 196                                                  u64 *storage_space,
 197                                                  u64 *remaining_space,
 198                                                  u64 *max_variable_size)
 199 {
 200         if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
 201                 return EFI_UNSUPPORTED;
 202
 203         return efi_call_virt4(query_variable_info, attr, storage_space,
 204                               remaining_space, max_variable_size);
 205 }
 206
 207 static efi_status_t virt_efi_get_next_high_mono_count(u32 *count)
 208 {
 209         return efi_call_virt1(get_next_high_mono_count, count);
 210 }
 211
 212 static void virt_efi_reset_system(int reset_type,
 213                                   efi_status_t status,
 214                                   unsigned long data_size,
 215                                   efi_char16_t *data)
 216 {
 217         efi_call_virt4(reset_system, reset_type, status,
 218                        data_size, data);
 219 }
 220
 221 static efi_status_t virt_efi_update_capsule(efi_capsule_header_t **capsules,
 222                                             unsigned long count,
 223                                             unsigned long sg_list)
 224 {
 225         if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
 226                 return EFI_UNSUPPORTED;
 227
 228         return efi_call_virt3(update_capsule, capsules, count, sg_list);
 229 }
 230
 231 static efi_status_t virt_efi_query_capsule_caps(efi_capsule_header_t **capsules,
 232                                                 unsigned long count,
 233                                                 u64 *max_size,
 234                                                 int *reset_type)
 235 {
 236         if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
 237                 return EFI_UNSUPPORTED;
 238
 239         return efi_call_virt4(query_capsule_caps, capsules, count, max_size,
 240                               reset_type);
 241 }
 242
 243 static efi_status_t __init phys_efi_set_virtual_address_map(
 244         unsigned long memory_map_size,
 245         unsigned long descriptor_size,
 246         u32 descriptor_version,
 247         efi_memory_desc_t *virtual_map)
 248 {
 249         efi_status_t status;
 250
 251         efi_call_phys_prelog();
 252         status = efi_call_phys4(efi_phys.set_virtual_address_map,
 253                                 memory_map_size, descriptor_size,
 254                                 descriptor_version, virtual_map);
 255         efi_call_phys_epilog();
 256         return status;
 257 }
 258
 259 static efi_status_t __init phys_efi_get_time(efi_time_t *tm,
 260                                              efi_time_cap_t *tc)
 261 {
 262         unsigned long flags;
 263         efi_status_t status;
 264
 265         spin_lock_irqsave(&rtc_lock, flags);
 266         efi_call_phys_prelog();
 267         status = efi_call_phys2(efi_phys.get_time, virt_to_phys(tm),
 268                                 virt_to_phys(tc));
 269         efi_call_phys_epilog();
 270         spin_unlock_irqrestore(&rtc_lock, flags);
 271         return status;
 272 }
 273
 274 int efi_set_rtc_mmss(const struct timespec *now)
 275 {
 276         unsigned long nowtime = now->tv_sec;
 277         efi_status_t    status;
 278         efi_time_t      eft;
 279         efi_time_cap_t  cap;
 280         struct rtc_time tm;
 281
 282         status = efi.get_time(&eft, &cap);
 283         if (status != EFI_SUCCESS) {
 284                 pr_err("Oops: efitime: can't read time!\n");
 285                 return -1;
 286         }
 287
 288         rtc_time_to_tm(nowtime, &tm);
 289         if (!rtc_valid_tm(&tm)) {
 290                 eft.year = tm.tm_year + 1900;
 291                 eft.month = tm.tm_mon + 1;
 292                 eft.day = tm.tm_mday;
 293                 eft.minute = tm.tm_min;
 294                 eft.second = tm.tm_sec;
 295                 eft.nanosecond = 0;
 296         } else {
 297                 printk(KERN_ERR
 298                        "%s: Invalid EFI RTC value: write of %lx to EFI RTC failed\n",
 299                        __FUNCTION__, nowtime);
 300                 return -1;
 301         }
 302
 303         status = efi.set_time(&eft);
 304         if (status != EFI_SUCCESS) {
 305                 pr_err("Oops: efitime: can't write time!\n");
 306                 return -1;
 307         }
 308         return 0;
 309 }
 310
 311 void efi_get_time(struct timespec *now)
 312 {
 313         efi_status_t status;
 314         efi_time_t eft;
 315         efi_time_cap_t cap;
 316
 317         status = efi.get_time(&eft, &cap);
 318         if (status != EFI_SUCCESS)
 319                 pr_err("Oops: efitime: can't read time!\n");
 320
 321         now->tv_sec = mktime(eft.year, eft.month, eft.day, eft.hour,
 322                              eft.minute, eft.second);
 323         now->tv_nsec = 0;
 324 }
 325
 326 /*
 327  * Tell the kernel about the EFI memory map.  This might include
 328  * more than the max 128 entries that can fit in the e820 legacy
 329  * (zeropage) memory map.
 330  */
 331
 332 static void __init do_add_efi_memmap(void)
 333 {
 334         void *p;
 335
 336         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 337                 efi_memory_desc_t *md = p;
 338                 unsigned long long start = md->phys_addr;
 339                 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
 340                 int e820_type;
 341
 342                 switch (md->type) {
 343                 case EFI_LOADER_CODE:
 344                 case EFI_LOADER_DATA:
 345                 case EFI_BOOT_SERVICES_CODE:
 346                 case EFI_BOOT_SERVICES_DATA:
 347                 case EFI_CONVENTIONAL_MEMORY:
 348                         if (md->attribute & EFI_MEMORY_WB)
 349                                 e820_type = E820_RAM;
 350                         else
 351                                 e820_type = E820_RESERVED;
 352                         break;
 353                 case EFI_ACPI_RECLAIM_MEMORY:
 354                         e820_type = E820_ACPI;
 355                         break;
 356                 case EFI_ACPI_MEMORY_NVS:
 357                         e820_type = E820_NVS;
 358                         break;
 359                 case EFI_UNUSABLE_MEMORY:
 360                         e820_type = E820_UNUSABLE;
 361                         break;
 362                 default:
 363                         /*
 364                          * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
 365                          * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
 366                          * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
 367                          */
 368                         e820_type = E820_RESERVED;
 369                         break;
 370                 }
 371                 e820_add_region(start, size, e820_type);
 372         }
 373         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 374 }
 375
 376 int __init efi_memblock_x86_reserve_range(void)
 377 {
 378         struct efi_info *e = &boot_params.efi_info;
 379         unsigned long pmap;
 380
 381 #ifdef CONFIG_X86_32
 382         /* Can't handle data above 4GB at this time */
 383         if (e->efi_memmap_hi) {
 384                 pr_err("Memory map is above 4GB, disabling EFI.\n");
 385                 return -EINVAL;
 386         }
 387         pmap =  e->efi_memmap;
 388 #else
 389         pmap = (e->efi_memmap | ((__u64)e->efi_memmap_hi << 32));
 390 #endif
 391         memmap.phys_map         = (void *)pmap;
 392         memmap.nr_map           = e->efi_memmap_size /
 393                                   e->efi_memdesc_size;
 394         memmap.desc_size        = e->efi_memdesc_size;
 395         memmap.desc_version     = e->efi_memdesc_version;
 396
 397         memblock_reserve(pmap, memmap.nr_map * memmap.desc_size);
 398
 399         efi.memmap = &memmap;
 400
 401         return 0;
 402 }
 403
 404 static void __init print_efi_memmap(void)
 405 {
 406 #ifdef EFI_DEBUG
 407         efi_memory_desc_t *md;
 408         void *p;
 409         int i;
 410
 411         for (p = memmap.map, i = 0;
 412              p < memmap.map_end;
 413              p += memmap.desc_size, i++) {
 414                 md = p;
 415                 pr_info("mem%02u: type=%u, attr=0x%llx, "
 416                         "range=[0x%016llx-0x%016llx) (%lluMB)\n",
 417                         i, md->type, md->attribute, md->phys_addr,
 418                         md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT),
 419                         (md->num_pages >> (20 - EFI_PAGE_SHIFT)));
 420         }
 421 #endif  /*  EFI_DEBUG  */
 422 }
 423
 424 void __init efi_reserve_boot_services(void)
 425 {
 426         void *p;
 427
 428         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 429                 efi_memory_desc_t *md = p;
 430                 u64 start = md->phys_addr;
 431                 u64 size = md->num_pages << EFI_PAGE_SHIFT;
 432
 433                 if (md->type != EFI_BOOT_SERVICES_CODE &&
 434                     md->type != EFI_BOOT_SERVICES_DATA)
 435                         continue;
 436                 /* Only reserve where possible:
 437                  * - Not within any already allocated areas
 438                  * - Not over any memory area (really needed, if above?)
 439                  * - Not within any part of the kernel
 440                  * - Not the bios reserved area
 441                 */
 442                 if ((start + size > __pa_symbol(_text)
 443                                 && start <= __pa_symbol(_end)) ||
 444                         !e820_all_mapped(start, start+size, E820_RAM) ||
 445                         memblock_is_region_reserved(start, size)) {
 446                         /* Could not reserve, skip it */
 447                         md->num_pages = 0;
 448                         memblock_dbg("Could not reserve boot range "
 449                                         "[0x%010llx-0x%010llx]\n",
 450                                                 start, start+size-1);
 451                 } else
 452                         memblock_reserve(start, size);
 453         }
 454 }
 455
 456 void __init efi_unmap_memmap(void)
 457 {
 458         clear_bit(EFI_MEMMAP, &x86_efi_facility);
 459         if (memmap.map) {
 460                 early_iounmap(memmap.map, memmap.nr_map * memmap.desc_size);
 461                 memmap.map = NULL;
 462         }
 463 }
 464
 465 void __init efi_free_boot_services(void)
 466 {
 467         void *p;
 468
 469         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 470                 efi_memory_desc_t *md = p;
 471                 unsigned long long start = md->phys_addr;
 472                 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
 473
 474                 if (md->type != EFI_BOOT_SERVICES_CODE &&
 475                     md->type != EFI_BOOT_SERVICES_DATA)
 476                         continue;
 477
 478                 /* Could not reserve boot area */
 479                 if (!size)
 480                         continue;
 481
 482                 free_bootmem_late(start, size);
 483         }
 484
 485         efi_unmap_memmap();
 486 }
 487
 488 static int __init efi_systab_init(void *phys)
 489 {
 490         if (efi_enabled(EFI_64BIT)) {
 491                 efi_system_table_64_t *systab64;
 492                 struct efi_setup_data *data = NULL;
 493                 u64 tmp = 0;
 494
 495                 if (efi_setup) {
 496                         data = early_memremap(efi_setup, sizeof(*data));
 497                         if (!data)
 498                                 return -ENOMEM;
 499                 }
 500                 systab64 = early_ioremap((unsigned long)phys,
 501                                          sizeof(*systab64));
 502                 if (systab64 == NULL) {
 503                         pr_err("Couldn't map the system table!\n");
 504                         if (data)
 505                                 early_iounmap(data, sizeof(*data));
 506                         return -ENOMEM;
 507                 }
 508
 509                 efi_systab.hdr = systab64->hdr;
 510                 efi_systab.fw_vendor = data ? (unsigned long)data->fw_vendor :
 511                                               systab64->fw_vendor;
 512                 tmp |= data ? data->fw_vendor : systab64->fw_vendor;
 513                 efi_systab.fw_revision = systab64->fw_revision;
 514                 efi_systab.con_in_handle = systab64->con_in_handle;
 515                 tmp |= systab64->con_in_handle;
 516                 efi_systab.con_in = systab64->con_in;
 517                 tmp |= systab64->con_in;
 518                 efi_systab.con_out_handle = systab64->con_out_handle;
 519                 tmp |= systab64->con_out_handle;
 520                 efi_systab.con_out = systab64->con_out;
 521                 tmp |= systab64->con_out;
 522                 efi_systab.stderr_handle = systab64->stderr_handle;
 523                 tmp |= systab64->stderr_handle;
 524                 efi_systab.stderr = systab64->stderr;
 525                 tmp |= systab64->stderr;
 526                 efi_systab.runtime = data ?
 527                                      (void *)(unsigned long)data->runtime :
 528                                      (void *)(unsigned long)systab64->runtime;
 529                 tmp |= data ? data->runtime : systab64->runtime;
 530                 efi_systab.boottime = (void *)(unsigned long)systab64->boottime;
 531                 tmp |= systab64->boottime;
 532                 efi_systab.nr_tables = systab64->nr_tables;
 533                 efi_systab.tables = data ? (unsigned long)data->tables :
 534                                            systab64->tables;
 535                 tmp |= data ? data->tables : systab64->tables;
 536
 537                 early_iounmap(systab64, sizeof(*systab64));
 538                 if (data)
 539                         early_iounmap(data, sizeof(*data));
 540 #ifdef CONFIG_X86_32
 541                 if (tmp >> 32) {
 542                         pr_err("EFI data located above 4GB, disabling EFI.\n");
 543                         return -EINVAL;
 544                 }
 545 #endif
 546         } else {
 547                 efi_system_table_32_t *systab32;
 548
 549                 systab32 = early_ioremap((unsigned long)phys,
 550                                          sizeof(*systab32));
 551                 if (systab32 == NULL) {
 552                         pr_err("Couldn't map the system table!\n");
 553                         return -ENOMEM;
 554                 }
 555
 556                 efi_systab.hdr = systab32->hdr;
 557                 efi_systab.fw_vendor = systab32->fw_vendor;
 558                 efi_systab.fw_revision = systab32->fw_revision;
 559                 efi_systab.con_in_handle = systab32->con_in_handle;
 560                 efi_systab.con_in = systab32->con_in;
 561                 efi_systab.con_out_handle = systab32->con_out_handle;
 562                 efi_systab.con_out = systab32->con_out;
 563                 efi_systab.stderr_handle = systab32->stderr_handle;
 564                 efi_systab.stderr = systab32->stderr;
 565                 efi_systab.runtime = (void *)(unsigned long)systab32->runtime;
 566                 efi_systab.boottime = (void *)(unsigned long)systab32->boottime;
 567                 efi_systab.nr_tables = systab32->nr_tables;
 568                 efi_systab.tables = systab32->tables;
 569
 570                 early_iounmap(systab32, sizeof(*systab32));
 571         }
 572
 573         efi.systab = &efi_systab;
 574
 575         /*
 576          * Verify the EFI Table
 577          */
 578         if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE) {
 579                 pr_err("System table signature incorrect!\n");
 580                 return -EINVAL;
 581         }
 582         if ((efi.systab->hdr.revision >> 16) == 0)
 583                 pr_err("Warning: System table version "
 584                        "%d.%02d, expected 1.00 or greater!\n",
 585                        efi.systab->hdr.revision >> 16,
 586                        efi.systab->hdr.revision & 0xffff);
 587
 588         return 0;
 589 }
 590
 591 static int __init efi_runtime_init(void)
 592 {
 593         efi_runtime_services_t *runtime;
 594
 595         /*
 596          * Check out the runtime services table. We need to map
 597          * the runtime services table so that we can grab the physical
 598          * address of several of the EFI runtime functions, needed to
 599          * set the firmware into virtual mode.
 600          */
 601         runtime = early_ioremap((unsigned long)efi.systab->runtime,
 602                                 sizeof(efi_runtime_services_t));
 603         if (!runtime) {
 604                 pr_err("Could not map the runtime service table!\n");
 605                 return -ENOMEM;
 606         }
 607         /*
 608          * We will only need *early* access to the following
 609          * two EFI runtime services before set_virtual_address_map
 610          * is invoked.
 611          */
 612         efi_phys.get_time = (efi_get_time_t *)runtime->get_time;
 613         efi_phys.set_virtual_address_map =
 614                 (efi_set_virtual_address_map_t *)
 615                 runtime->set_virtual_address_map;
 616         /*
 617          * Make efi_get_time can be called before entering
 618          * virtual mode.
 619          */
 620         efi.get_time = phys_efi_get_time;
 621         early_iounmap(runtime, sizeof(efi_runtime_services_t));
 622
 623         return 0;
 624 }
 625
 626 static int __init efi_memmap_init(void)
 627 {
 628         /* Map the EFI memory map */
 629         memmap.map = early_ioremap((unsigned long)memmap.phys_map,
 630                                    memmap.nr_map * memmap.desc_size);
 631         if (memmap.map == NULL) {
 632                 pr_err("Could not map the memory map!\n");
 633                 return -ENOMEM;
 634         }
 635         memmap.map_end = memmap.map + (memmap.nr_map * memmap.desc_size);
 636
 637         if (add_efi_memmap)
 638                 do_add_efi_memmap();
 639
 640         return 0;
 641 }
 642
 643 /*
 644  * A number of config table entries get remapped to virtual addresses
 645  * after entering EFI virtual mode. However, the kexec kernel requires
 646  * their physical addresses therefore we pass them via setup_data and
 647  * correct those entries to their respective physical addresses here.
 648  *
 649  * Currently only handles smbios which is necessary for some firmware
 650  * implementation.
 651  */
 652 static int __init efi_reuse_config(u64 tables, int nr_tables)
 653 {
 654         int i, sz, ret = 0;
 655         void *p, *tablep;
 656         struct efi_setup_data *data;
 657
 658         if (!efi_setup)
 659                 return 0;
 660
 661         if (!efi_enabled(EFI_64BIT))
 662                 return 0;
 663
 664         data = early_memremap(efi_setup, sizeof(*data));
 665         if (!data) {
 666                 ret = -ENOMEM;
 667                 goto out;
 668         }
 669
 670         if (!data->smbios)
 671                 goto out_memremap;
 672
 673         sz = sizeof(efi_config_table_64_t);
 674
 675         p = tablep = early_memremap(tables, nr_tables * sz);
 676         if (!p) {
 677                 pr_err("Could not map Configuration table!\n");
 678                 ret = -ENOMEM;
 679                 goto out_memremap;
 680         }
 681
 682         for (i = 0; i < efi.systab->nr_tables; i++) {
 683                 efi_guid_t guid;
 684
 685                 guid = ((efi_config_table_64_t *)p)->guid;
 686
 687                 if (!efi_guidcmp(guid, SMBIOS_TABLE_GUID))
 688                         ((efi_config_table_64_t *)p)->table = data->smbios;
 689                 p += sz;
 690         }
 691         early_iounmap(tablep, nr_tables * sz);
 692
 693 out_memremap:
 694         early_iounmap(data, sizeof(*data));
 695 out:
 696         return ret;
 697 }
 698
 699 void __init efi_init(void)
 700 {
 701         efi_char16_t *c16;
 702         char vendor[100] = "unknown";
 703         int i = 0;
 704         void *tmp;
 705
 706 #ifdef CONFIG_X86_32
 707         if (boot_params.efi_info.efi_systab_hi ||
 708             boot_params.efi_info.efi_memmap_hi) {
 709                 pr_info("Table located above 4GB, disabling EFI.\n");
 710                 return;
 711         }
 712         efi_phys.systab = (efi_system_table_t *)boot_params.efi_info.efi_systab;
 713 #else
 714         efi_phys.systab = (efi_system_table_t *)
 715                           (boot_params.efi_info.efi_systab |
 716                           ((__u64)boot_params.efi_info.efi_systab_hi<<32));
 717 #endif
 718
 719         if (efi_systab_init(efi_phys.systab))
 720                 return;
 721
 722         set_bit(EFI_SYSTEM_TABLES, &x86_efi_facility);
 723
 724         efi.config_table = (unsigned long)efi.systab->tables;
 725         efi.fw_vendor    = (unsigned long)efi.systab->fw_vendor;
 726         efi.runtime      = (unsigned long)efi.systab->runtime;
 727
 728         /*
 729          * Show what we know for posterity
 730          */
 731         c16 = tmp = early_ioremap(efi.systab->fw_vendor, 2);
 732         if (c16) {
 733                 for (i = 0; i < sizeof(vendor) - 1 && *c16; ++i)
 734                         vendor[i] = *c16++;
 735                 vendor[i] = '\0';
 736         } else
 737                 pr_err("Could not map the firmware vendor!\n");
 738         early_iounmap(tmp, 2);
 739
 740         pr_info("EFI v%u.%.02u by %s\n",
 741                 efi.systab->hdr.revision >> 16,
 742                 efi.systab->hdr.revision & 0xffff, vendor);
 743
 744         if (efi_reuse_config(efi.systab->tables, efi.systab->nr_tables))
 745                 return;
 746
 747         if (efi_config_init(arch_tables))
 748                 return;
 749
 750         set_bit(EFI_CONFIG_TABLES, &x86_efi_facility);
 751
 752         /*
 753          * Note: We currently don't support runtime services on an EFI
 754          * that doesn't match the kernel 32/64-bit mode.
 755          */
 756
 757         if (!efi_is_native())
 758                 pr_info("No EFI runtime due to 32/64-bit mismatch with kernel\n");
 759         else {
 760                 if (disable_runtime || efi_runtime_init())
 761                         return;
 762                 set_bit(EFI_RUNTIME_SERVICES, &x86_efi_facility);
 763         }
 764         if (efi_memmap_init())
 765                 return;
 766
 767         set_bit(EFI_MEMMAP, &x86_efi_facility);
 768
 769         print_efi_memmap();
 770 }
 771
 772 void __init efi_late_init(void)
 773 {
 774         efi_bgrt_init();
 775 }
 776
 777 void __init efi_set_executable(efi_memory_desc_t *md, bool executable)
 778 {
 779         u64 addr, npages;
 780
 781         addr = md->virt_addr;
 782         npages = md->num_pages;
 783
 784         memrange_efi_to_native(&addr, &npages);
 785
 786         if (executable)
 787                 set_memory_x(addr, npages);
 788         else
 789                 set_memory_nx(addr, npages);
 790 }
 791
 792 void __init runtime_code_page_mkexec(void)
 793 {
 794         efi_memory_desc_t *md;
 795         void *p;
 796
 797         /* Make EFI runtime service code area executable */
 798         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 799                 md = p;
 800
 801                 if (md->type != EFI_RUNTIME_SERVICES_CODE)
 802                         continue;
 803
 804                 efi_set_executable(md, true);
 805         }
 806 }
 807
 808 void efi_memory_uc(u64 addr, unsigned long size)
 809 {
 810         unsigned long page_shift = 1UL << EFI_PAGE_SHIFT;
 811         u64 npages;
 812
 813         npages = round_up(size, page_shift) / page_shift;
 814         memrange_efi_to_native(&addr, &npages);
 815         set_memory_uc(addr, npages);
 816 }
 817
 818 void __init old_map_region(efi_memory_desc_t *md)
 819 {
 820         u64 start_pfn, end_pfn, end;
 821         unsigned long size;
 822         void *va;
 823
 824         start_pfn = PFN_DOWN(md->phys_addr);
 825         size      = md->num_pages << PAGE_SHIFT;
 826         end       = md->phys_addr + size;
 827         end_pfn   = PFN_UP(end);
 828
 829         if (pfn_range_is_mapped(start_pfn, end_pfn)) {
 830                 va = __va(md->phys_addr);
 831
 832                 if (!(md->attribute & EFI_MEMORY_WB))
 833                         efi_memory_uc((u64)(unsigned long)va, size);
 834         } else
 835                 va = efi_ioremap(md->phys_addr, size,
 836                                  md->type, md->attribute);
 837
 838         md->virt_addr = (u64) (unsigned long) va;
 839         if (!va)
 840                 pr_err("ioremap of 0x%llX failed!\n",
 841                        (unsigned long long)md->phys_addr);
 842 }
 843
 844 /* Merge contiguous regions of the same type and attribute */
 845 static void __init efi_merge_regions(void)
 846 {
 847         void *p;
 848         efi_memory_desc_t *md, *prev_md = NULL;
 849
 850         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 851                 u64 prev_size;
 852                 md = p;
 853
 854                 if (!prev_md) {
 855                         prev_md = md;
 856                         continue;
 857                 }
 858
 859                 if (prev_md->type != md->type ||
 860                     prev_md->attribute != md->attribute) {
 861                         prev_md = md;
 862                         continue;
 863                 }
 864
 865                 prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
 866
 867                 if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
 868                         prev_md->num_pages += md->num_pages;
 869                         md->type = EFI_RESERVED_TYPE;
 870                         md->attribute = 0;
 871                         continue;
 872                 }
 873                 prev_md = md;
 874         }
 875 }
 876
 877 static void __init get_systab_virt_addr(efi_memory_desc_t *md)
 878 {
 879         unsigned long size;
 880         u64 end, systab;
 881
 882         size = md->num_pages << EFI_PAGE_SHIFT;
 883         end = md->phys_addr + size;
 884         systab = (u64)(unsigned long)efi_phys.systab;
 885         if (md->phys_addr <= systab && systab < end) {
 886                 systab += md->virt_addr - md->phys_addr;
 887                 efi.systab = (efi_system_table_t *)(unsigned long)systab;
 888         }
 889 }
 890
 891 static int __init save_runtime_map(void)
 892 {
 893         efi_memory_desc_t *md;
 894         void *tmp, *p, *q = NULL;
 895         int count = 0;
 896
 897         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 898                 md = p;
 899
 900                 if (!(md->attribute & EFI_MEMORY_RUNTIME) ||
 901                     (md->type == EFI_BOOT_SERVICES_CODE) ||
 902                     (md->type == EFI_BOOT_SERVICES_DATA))
 903                         continue;
 904                 tmp = krealloc(q, (count + 1) * memmap.desc_size, GFP_KERNEL);
 905                 if (!tmp)
 906                         goto out;
 907                 q = tmp;
 908
 909                 memcpy(q + count * memmap.desc_size, md, memmap.desc_size);
 910                 count++;
 911         }
 912
 913         efi_runtime_map_setup(q, count, memmap.desc_size);
 914
 915         return 0;
 916 out:
 917         kfree(q);
 918         return -ENOMEM;
 919 }
 920
 921 /*
 922  * Map efi regions which were passed via setup_data. The virt_addr is a fixed
 923  * addr which was used in first kernel of a kexec boot.
 924  */
 925 static void __init efi_map_regions_fixed(void)
 926 {
 927         void *p;
 928         efi_memory_desc_t *md;
 929
 930         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 931                 md = p;
 932                 efi_map_region_fixed(md); /* FIXME: add error handling */
 933                 get_systab_virt_addr(md);
 934         }
 935
 936 }
 937
 938 /*
 939  * Map efi memory ranges for runtime serivce and update new_memmap with virtual
 940  * addresses.
 941  */
 942 static void * __init efi_map_regions(int *count)
 943 {
 944         efi_memory_desc_t *md;
 945         void *p, *tmp, *new_memmap = NULL;
 946
 947         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 948                 md = p;
 949                 if (!(md->attribute & EFI_MEMORY_RUNTIME)) {
 950 #ifdef CONFIG_X86_64
 951                         if (md->type != EFI_BOOT_SERVICES_CODE &&
 952                             md->type != EFI_BOOT_SERVICES_DATA)
 953 #endif
 954                                 continue;
 955                 }
 956
 957                 efi_map_region(md);
 958                 get_systab_virt_addr(md);
 959
 960                 tmp = krealloc(new_memmap, (*count + 1) * memmap.desc_size,
 961                                GFP_KERNEL);
 962                 if (!tmp)
 963                         goto out;
 964                 new_memmap = tmp;
 965                 memcpy(new_memmap + (*count * memmap.desc_size), md,
 966                        memmap.desc_size);
 967                 (*count)++;
 968         }
 969
 970         return new_memmap;
 971 out:
 972         kfree(new_memmap);
 973         return NULL;
 974 }
 975
 976 /*
 977  * This function will switch the EFI runtime services to virtual mode.
 978  * Essentially, we look through the EFI memmap and map every region that
 979  * has the runtime attribute bit set in its memory descriptor into the
 980  * ->trampoline_pgd page table using a top-down VA allocation scheme.
 981  *
 982  * The old method which used to update that memory descriptor with the
 983  * virtual address obtained from ioremap() is still supported when the
 984  * kernel is booted with efi=old_map on its command line. Same old
 985  * method enabled the runtime services to be called without having to
 986  * thunk back into physical mode for every invocation.
 987  *
 988  * The new method does a pagetable switch in a preemption-safe manner
 989  * so that we're in a different address space when calling a runtime
 990  * function. For function arguments passing we do copy the PGDs of the
 991  * kernel page table into ->trampoline_pgd prior to each call.
 992  *
 993  * Specially for kexec boot, efi runtime maps in previous kernel should
 994  * be passed in via setup_data. In that case runtime ranges will be mapped
 995  * to the same virtual addresses as the first kernel.
 996  */
 997 void __init efi_enter_virtual_mode(void)
 998 {
 999         efi_status_t status;
1000         void *new_memmap = NULL;
1001         int err, count = 0;
1002
1003         efi.systab = NULL;
1004
1005         if (efi_setup) {
1006                 efi_map_regions_fixed();
1007         } else {
1008                 efi_merge_regions();
1009                 new_memmap = efi_map_regions(&count);
1010                 if (!new_memmap) {
1011                         pr_err("Error reallocating memory, EFI runtime non-functional!\n");
1012                         return;
1013                 }
1014         }
1015
1016         err = save_runtime_map();
1017         if (err)
1018                 pr_err("Error saving runtime map, efi runtime on kexec non-functional!!\n");
1019
1020         BUG_ON(!efi.systab);
1021
1022         efi_setup_page_tables();
1023         efi_sync_low_kernel_mappings();
1024
1025         if (!efi_setup) {
1026                 status = phys_efi_set_virtual_address_map(
1027                         memmap.desc_size * count,
1028                         memmap.desc_size,
1029                         memmap.desc_version,
1030                         (efi_memory_desc_t *)__pa(new_memmap));
1031
1032                 if (status != EFI_SUCCESS) {
1033                         pr_alert("Unable to switch EFI into virtual mode (status=%lx)!\n",
1034                                  status);
1035                         panic("EFI call to SetVirtualAddressMap() failed!");
1036                 }
1037         }
1038
1039         /*
1040          * Now that EFI is in virtual mode, update the function
1041          * pointers in the runtime service table to the new virtual addresses.
1042          *
1043          * Call EFI services through wrapper functions.
1044          */
1045         efi.runtime_version = efi_systab.hdr.revision;
1046         efi.get_time = virt_efi_get_time;
1047         efi.set_time = virt_efi_set_time;
1048         efi.get_wakeup_time = virt_efi_get_wakeup_time;
1049         efi.set_wakeup_time = virt_efi_set_wakeup_time;
1050         efi.get_variable = virt_efi_get_variable;
1051         efi.get_next_variable = virt_efi_get_next_variable;
1052         efi.set_variable = virt_efi_set_variable;
1053         efi.get_next_high_mono_count = virt_efi_get_next_high_mono_count;
1054         efi.reset_system = virt_efi_reset_system;
1055         efi.set_virtual_address_map = NULL;
1056         efi.query_variable_info = virt_efi_query_variable_info;
1057         efi.update_capsule = virt_efi_update_capsule;
1058         efi.query_capsule_caps = virt_efi_query_capsule_caps;
1059
1060         efi_runtime_mkexec();
1061
1062         kfree(new_memmap);
1063
1064         /* clean DUMMY object */
1065         efi.set_variable(efi_dummy_name, &EFI_DUMMY_GUID,
1066                          EFI_VARIABLE_NON_VOLATILE |
1067                          EFI_VARIABLE_BOOTSERVICE_ACCESS |
1068                          EFI_VARIABLE_RUNTIME_ACCESS,
1069                          0, NULL);
1070 }
1071
1072 /*
1073  * Convenience functions to obtain memory types and attributes
1074  */
1075 u32 efi_mem_type(unsigned long phys_addr)
1076 {
1077         efi_memory_desc_t *md;
1078         void *p;
1079
1080         if (!efi_enabled(EFI_MEMMAP))
1081                 return 0;
1082
1083         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
1084                 md = p;
1085                 if ((md->phys_addr <= phys_addr) &&
1086                     (phys_addr < (md->phys_addr +
1087                                   (md->num_pages << EFI_PAGE_SHIFT))))
1088                         return md->type;
1089         }
1090         return 0;
1091 }
1092
1093 u64 efi_mem_attributes(unsigned long phys_addr)
1094 {
1095         efi_memory_desc_t *md;
1096         void *p;
1097
1098         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
1099                 md = p;
1100                 if ((md->phys_addr <= phys_addr) &&
1101                     (phys_addr < (md->phys_addr +
1102                                   (md->num_pages << EFI_PAGE_SHIFT))))
1103                         return md->attribute;
1104         }
1105         return 0;
1106 }
1107
1108 /*
1109  * Some firmware has serious problems when using more than 50% of the EFI
1110  * variable store, i.e. it triggers bugs that can brick machines. Ensure that
1111  * we never use more than this safe limit.
1112  *
1113  * Return EFI_SUCCESS if it is safe to write 'size' bytes to the variable
1114  * store.
1115  */
1116 efi_status_t efi_query_variable_store(u32 attributes, unsigned long size)
1117 {
1118         efi_status_t status;
1119         u64 storage_size, remaining_size, max_size;
1120
1121         if (!(attributes & EFI_VARIABLE_NON_VOLATILE))
1122                 return 0;
1123
1124         status = efi.query_variable_info(attributes, &storage_size,
1125                                          &remaining_size, &max_size);
1126         if (status != EFI_SUCCESS)
1127                 return status;
1128
1129         /*
1130          * Some firmware implementations refuse to boot if there's insufficient
1131          * space in the variable store. We account for that by refusing the
1132          * write if permitting it would reduce the available space to under
1133          * 5KB. This figure was provided by Samsung, so should be safe.
1134          */
1135         if ((remaining_size - size < EFI_MIN_RESERVE) &&
1136                 !efi_no_storage_paranoia) {
1137
1138                 /*
1139                  * Triggering garbage collection may require that the firmware
1140                  * generate a real EFI_OUT_OF_RESOURCES error. We can force
1141                  * that by attempting to use more space than is available.
1142                  */
1143                 unsigned long dummy_size = remaining_size + 1024;
1144                 void *dummy = kzalloc(dummy_size, GFP_ATOMIC);
1145
1146                 if (!dummy)
1147                         return EFI_OUT_OF_RESOURCES;
1148
1149                 status = efi.set_variable(efi_dummy_name, &EFI_DUMMY_GUID,
1150                                           EFI_VARIABLE_NON_VOLATILE |
1151                                           EFI_VARIABLE_BOOTSERVICE_ACCESS |
1152                                           EFI_VARIABLE_RUNTIME_ACCESS,
1153                                           dummy_size, dummy);
1154
1155                 if (status == EFI_SUCCESS) {
1156                         /*
1157                          * This should have failed, so if it didn't make sure
1158                          * that we delete it...
1159                          */
1160                         efi.set_variable(efi_dummy_name, &EFI_DUMMY_GUID,
1161                                          EFI_VARIABLE_NON_VOLATILE |
1162                                          EFI_VARIABLE_BOOTSERVICE_ACCESS |
1163                                          EFI_VARIABLE_RUNTIME_ACCESS,
1164                                          0, dummy);
1165                 }
1166
1167                 kfree(dummy);
1168
1169                 /*
1170                  * The runtime code may now have triggered a garbage collection
1171                  * run, so check the variable info again
1172                  */
1173                 status = efi.query_variable_info(attributes, &storage_size,
1174                                                  &remaining_size, &max_size);
1175
1176                 if (status != EFI_SUCCESS)
1177                         return status;
1178
1179                 /*
1180                  * There still isn't enough room, so return an error
1181                  */
1182                 if (remaining_size - size < EFI_MIN_RESERVE)
1183                         return EFI_OUT_OF_RESOURCES;
1184         }
1185
1186         return EFI_SUCCESS;
1187 }
1188 EXPORT_SYMBOL_GPL(efi_query_variable_store);
1189
1190 static int __init parse_efi_cmdline(char *str)
1191 {
1192         if (*str == '=')
1193                 str++;
1194
1195         if (!strncmp(str, "old_map", 7))
1196                 set_bit(EFI_OLD_MEMMAP, &x86_efi_facility);
1197
1198         return 0;
1199 }
1200 early_param("efi", parse_efi_cmdline);