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 static __initdata efi_config_table_type_t arch_tables[] = {
  71 #ifdef CONFIG_X86_UV
  72         {UV_SYSTEM_TABLE_GUID, "UVsystab", &efi.uv_systab},
  73 #endif
  74         {NULL_GUID, NULL, NULL},
  75 };
  76
  77 u64 efi_setup;          /* efi setup_data physical address */
  78
  79 static bool __initdata disable_runtime = false;
  80 static int __init setup_noefi(char *arg)
  81 {
  82         disable_runtime = true;
  83         return 0;
  84 }
  85 early_param("noefi", setup_noefi);
  86
  87 int add_efi_memmap;
  88 EXPORT_SYMBOL(add_efi_memmap);
  89
  90 static int __init setup_add_efi_memmap(char *arg)
  91 {
  92         add_efi_memmap = 1;
  93         return 0;
  94 }
  95 early_param("add_efi_memmap", setup_add_efi_memmap);
  96
  97 static bool efi_no_storage_paranoia;
  98
  99 static int __init setup_storage_paranoia(char *arg)
 100 {
 101         efi_no_storage_paranoia = true;
 102         return 0;
 103 }
 104 early_param("efi_no_storage_paranoia", setup_storage_paranoia);
 105
 106 static efi_status_t virt_efi_get_time(efi_time_t *tm, efi_time_cap_t *tc)
 107 {
 108         unsigned long flags;
 109         efi_status_t status;
 110
 111         spin_lock_irqsave(&rtc_lock, flags);
 112         status = efi_call_virt2(get_time, tm, tc);
 113         spin_unlock_irqrestore(&rtc_lock, flags);
 114         return status;
 115 }
 116
 117 static efi_status_t virt_efi_set_time(efi_time_t *tm)
 118 {
 119         unsigned long flags;
 120         efi_status_t status;
 121
 122         spin_lock_irqsave(&rtc_lock, flags);
 123         status = efi_call_virt1(set_time, tm);
 124         spin_unlock_irqrestore(&rtc_lock, flags);
 125         return status;
 126 }
 127
 128 static efi_status_t virt_efi_get_wakeup_time(efi_bool_t *enabled,
 129                                              efi_bool_t *pending,
 130                                              efi_time_t *tm)
 131 {
 132         unsigned long flags;
 133         efi_status_t status;
 134
 135         spin_lock_irqsave(&rtc_lock, flags);
 136         status = efi_call_virt3(get_wakeup_time,
 137                                 enabled, pending, tm);
 138         spin_unlock_irqrestore(&rtc_lock, flags);
 139         return status;
 140 }
 141
 142 static efi_status_t virt_efi_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm)
 143 {
 144         unsigned long flags;
 145         efi_status_t status;
 146
 147         spin_lock_irqsave(&rtc_lock, flags);
 148         status = efi_call_virt2(set_wakeup_time,
 149                                 enabled, tm);
 150         spin_unlock_irqrestore(&rtc_lock, flags);
 151         return status;
 152 }
 153
 154 static efi_status_t virt_efi_get_variable(efi_char16_t *name,
 155                                           efi_guid_t *vendor,
 156                                           u32 *attr,
 157                                           unsigned long *data_size,
 158                                           void *data)
 159 {
 160         return efi_call_virt5(get_variable,
 161                               name, vendor, attr,
 162                               data_size, data);
 163 }
 164
 165 static efi_status_t virt_efi_get_next_variable(unsigned long *name_size,
 166                                                efi_char16_t *name,
 167                                                efi_guid_t *vendor)
 168 {
 169         return efi_call_virt3(get_next_variable,
 170                               name_size, name, vendor);
 171 }
 172
 173 static efi_status_t virt_efi_set_variable(efi_char16_t *name,
 174                                           efi_guid_t *vendor,
 175                                           u32 attr,
 176                                           unsigned long data_size,
 177                                           void *data)
 178 {
 179         return efi_call_virt5(set_variable,
 180                               name, vendor, attr,
 181                               data_size, data);
 182 }
 183
 184 static efi_status_t virt_efi_query_variable_info(u32 attr,
 185                                                  u64 *storage_space,
 186                                                  u64 *remaining_space,
 187                                                  u64 *max_variable_size)
 188 {
 189         if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
 190                 return EFI_UNSUPPORTED;
 191
 192         return efi_call_virt4(query_variable_info, attr, storage_space,
 193                               remaining_space, max_variable_size);
 194 }
 195
 196 static efi_status_t virt_efi_get_next_high_mono_count(u32 *count)
 197 {
 198         return efi_call_virt1(get_next_high_mono_count, count);
 199 }
 200
 201 static void virt_efi_reset_system(int reset_type,
 202                                   efi_status_t status,
 203                                   unsigned long data_size,
 204                                   efi_char16_t *data)
 205 {
 206         efi_call_virt4(reset_system, reset_type, status,
 207                        data_size, data);
 208 }
 209
 210 static efi_status_t virt_efi_update_capsule(efi_capsule_header_t **capsules,
 211                                             unsigned long count,
 212                                             unsigned long sg_list)
 213 {
 214         if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
 215                 return EFI_UNSUPPORTED;
 216
 217         return efi_call_virt3(update_capsule, capsules, count, sg_list);
 218 }
 219
 220 static efi_status_t virt_efi_query_capsule_caps(efi_capsule_header_t **capsules,
 221                                                 unsigned long count,
 222                                                 u64 *max_size,
 223                                                 int *reset_type)
 224 {
 225         if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
 226                 return EFI_UNSUPPORTED;
 227
 228         return efi_call_virt4(query_capsule_caps, capsules, count, max_size,
 229                               reset_type);
 230 }
 231
 232 static efi_status_t __init phys_efi_set_virtual_address_map(
 233         unsigned long memory_map_size,
 234         unsigned long descriptor_size,
 235         u32 descriptor_version,
 236         efi_memory_desc_t *virtual_map)
 237 {
 238         efi_status_t status;
 239
 240         efi_call_phys_prelog();
 241         status = efi_call_phys4(efi_phys.set_virtual_address_map,
 242                                 memory_map_size, descriptor_size,
 243                                 descriptor_version, virtual_map);
 244         efi_call_phys_epilog();
 245         return status;
 246 }
 247
 248 static efi_status_t __init phys_efi_get_time(efi_time_t *tm,
 249                                              efi_time_cap_t *tc)
 250 {
 251         unsigned long flags;
 252         efi_status_t status;
 253
 254         spin_lock_irqsave(&rtc_lock, flags);
 255         efi_call_phys_prelog();
 256         status = efi_call_phys2(efi_phys.get_time, virt_to_phys(tm),
 257                                 virt_to_phys(tc));
 258         efi_call_phys_epilog();
 259         spin_unlock_irqrestore(&rtc_lock, flags);
 260         return status;
 261 }
 262
 263 int efi_set_rtc_mmss(const struct timespec *now)
 264 {
 265         unsigned long nowtime = now->tv_sec;
 266         efi_status_t    status;
 267         efi_time_t      eft;
 268         efi_time_cap_t  cap;
 269         struct rtc_time tm;
 270
 271         status = efi.get_time(&eft, &cap);
 272         if (status != EFI_SUCCESS) {
 273                 pr_err("Oops: efitime: can't read time!\n");
 274                 return -1;
 275         }
 276
 277         rtc_time_to_tm(nowtime, &tm);
 278         if (!rtc_valid_tm(&tm)) {
 279                 eft.year = tm.tm_year + 1900;
 280                 eft.month = tm.tm_mon + 1;
 281                 eft.day = tm.tm_mday;
 282                 eft.minute = tm.tm_min;
 283                 eft.second = tm.tm_sec;
 284                 eft.nanosecond = 0;
 285         } else {
 286                 printk(KERN_ERR
 287                        "%s: Invalid EFI RTC value: write of %lx to EFI RTC failed\n",
 288                        __FUNCTION__, nowtime);
 289                 return -1;
 290         }
 291
 292         status = efi.set_time(&eft);
 293         if (status != EFI_SUCCESS) {
 294                 pr_err("Oops: efitime: can't write time!\n");
 295                 return -1;
 296         }
 297         return 0;
 298 }
 299
 300 void efi_get_time(struct timespec *now)
 301 {
 302         efi_status_t status;
 303         efi_time_t eft;
 304         efi_time_cap_t cap;
 305
 306         status = efi.get_time(&eft, &cap);
 307         if (status != EFI_SUCCESS)
 308                 pr_err("Oops: efitime: can't read time!\n");
 309
 310         now->tv_sec = mktime(eft.year, eft.month, eft.day, eft.hour,
 311                              eft.minute, eft.second);
 312         now->tv_nsec = 0;
 313 }
 314
 315 /*
 316  * Tell the kernel about the EFI memory map.  This might include
 317  * more than the max 128 entries that can fit in the e820 legacy
 318  * (zeropage) memory map.
 319  */
 320
 321 static void __init do_add_efi_memmap(void)
 322 {
 323         void *p;
 324
 325         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 326                 efi_memory_desc_t *md = p;
 327                 unsigned long long start = md->phys_addr;
 328                 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
 329                 int e820_type;
 330
 331                 switch (md->type) {
 332                 case EFI_LOADER_CODE:
 333                 case EFI_LOADER_DATA:
 334                 case EFI_BOOT_SERVICES_CODE:
 335                 case EFI_BOOT_SERVICES_DATA:
 336                 case EFI_CONVENTIONAL_MEMORY:
 337                         if (md->attribute & EFI_MEMORY_WB)
 338                                 e820_type = E820_RAM;
 339                         else
 340                                 e820_type = E820_RESERVED;
 341                         break;
 342                 case EFI_ACPI_RECLAIM_MEMORY:
 343                         e820_type = E820_ACPI;
 344                         break;
 345                 case EFI_ACPI_MEMORY_NVS:
 346                         e820_type = E820_NVS;
 347                         break;
 348                 case EFI_UNUSABLE_MEMORY:
 349                         e820_type = E820_UNUSABLE;
 350                         break;
 351                 default:
 352                         /*
 353                          * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
 354                          * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
 355                          * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
 356                          */
 357                         e820_type = E820_RESERVED;
 358                         break;
 359                 }
 360                 e820_add_region(start, size, e820_type);
 361         }
 362         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 363 }
 364
 365 int __init efi_memblock_x86_reserve_range(void)
 366 {
 367         struct efi_info *e = &boot_params.efi_info;
 368         unsigned long pmap;
 369
 370 #ifdef CONFIG_X86_32
 371         /* Can't handle data above 4GB at this time */
 372         if (e->efi_memmap_hi) {
 373                 pr_err("Memory map is above 4GB, disabling EFI.\n");
 374                 return -EINVAL;
 375         }
 376         pmap =  e->efi_memmap;
 377 #else
 378         pmap = (e->efi_memmap | ((__u64)e->efi_memmap_hi << 32));
 379 #endif
 380         memmap.phys_map         = (void *)pmap;
 381         memmap.nr_map           = e->efi_memmap_size /
 382                                   e->efi_memdesc_size;
 383         memmap.desc_size        = e->efi_memdesc_size;
 384         memmap.desc_version     = e->efi_memdesc_version;
 385
 386         memblock_reserve(pmap, memmap.nr_map * memmap.desc_size);
 387
 388         efi.memmap = &memmap;
 389
 390         return 0;
 391 }
 392
 393 static void __init print_efi_memmap(void)
 394 {
 395 #ifdef EFI_DEBUG
 396         efi_memory_desc_t *md;
 397         void *p;
 398         int i;
 399
 400         for (p = memmap.map, i = 0;
 401              p < memmap.map_end;
 402              p += memmap.desc_size, i++) {
 403                 md = p;
 404                 pr_info("mem%02u: type=%u, attr=0x%llx, "
 405                         "range=[0x%016llx-0x%016llx) (%lluMB)\n",
 406                         i, md->type, md->attribute, md->phys_addr,
 407                         md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT),
 408                         (md->num_pages >> (20 - EFI_PAGE_SHIFT)));
 409         }
 410 #endif  /*  EFI_DEBUG  */
 411 }
 412
 413 void __init efi_reserve_boot_services(void)
 414 {
 415         void *p;
 416
 417         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 418                 efi_memory_desc_t *md = p;
 419                 u64 start = md->phys_addr;
 420                 u64 size = md->num_pages << EFI_PAGE_SHIFT;
 421
 422                 if (md->type != EFI_BOOT_SERVICES_CODE &&
 423                     md->type != EFI_BOOT_SERVICES_DATA)
 424                         continue;
 425                 /* Only reserve where possible:
 426                  * - Not within any already allocated areas
 427                  * - Not over any memory area (really needed, if above?)
 428                  * - Not within any part of the kernel
 429                  * - Not the bios reserved area
 430                 */
 431                 if ((start + size > __pa_symbol(_text)
 432                                 && start <= __pa_symbol(_end)) ||
 433                         !e820_all_mapped(start, start+size, E820_RAM) ||
 434                         memblock_is_region_reserved(start, size)) {
 435                         /* Could not reserve, skip it */
 436                         md->num_pages = 0;
 437                         memblock_dbg("Could not reserve boot range "
 438                                         "[0x%010llx-0x%010llx]\n",
 439                                                 start, start+size-1);
 440                 } else
 441                         memblock_reserve(start, size);
 442         }
 443 }
 444
 445 void __init efi_unmap_memmap(void)
 446 {
 447         clear_bit(EFI_MEMMAP, &efi.flags);
 448         if (memmap.map) {
 449                 early_iounmap(memmap.map, memmap.nr_map * memmap.desc_size);
 450                 memmap.map = NULL;
 451         }
 452 }
 453
 454 void __init efi_free_boot_services(void)
 455 {
 456         void *p;
 457
 458         if (!efi_is_native())
 459                 return;
 460
 461         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 462                 efi_memory_desc_t *md = p;
 463                 unsigned long long start = md->phys_addr;
 464                 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
 465
 466                 if (md->type != EFI_BOOT_SERVICES_CODE &&
 467                     md->type != EFI_BOOT_SERVICES_DATA)
 468                         continue;
 469
 470                 /* Could not reserve boot area */
 471                 if (!size)
 472                         continue;
 473
 474                 free_bootmem_late(start, size);
 475         }
 476
 477         efi_unmap_memmap();
 478 }
 479
 480 static int __init efi_systab_init(void *phys)
 481 {
 482         if (efi_enabled(EFI_64BIT)) {
 483                 efi_system_table_64_t *systab64;
 484                 struct efi_setup_data *data = NULL;
 485                 u64 tmp = 0;
 486
 487                 if (efi_setup) {
 488                         data = early_memremap(efi_setup, sizeof(*data));
 489                         if (!data)
 490                                 return -ENOMEM;
 491                 }
 492                 systab64 = early_ioremap((unsigned long)phys,
 493                                          sizeof(*systab64));
 494                 if (systab64 == NULL) {
 495                         pr_err("Couldn't map the system table!\n");
 496                         if (data)
 497                                 early_iounmap(data, sizeof(*data));
 498                         return -ENOMEM;
 499                 }
 500
 501                 efi_systab.hdr = systab64->hdr;
 502                 efi_systab.fw_vendor = data ? (unsigned long)data->fw_vendor :
 503                                               systab64->fw_vendor;
 504                 tmp |= data ? data->fw_vendor : systab64->fw_vendor;
 505                 efi_systab.fw_revision = systab64->fw_revision;
 506                 efi_systab.con_in_handle = systab64->con_in_handle;
 507                 tmp |= systab64->con_in_handle;
 508                 efi_systab.con_in = systab64->con_in;
 509                 tmp |= systab64->con_in;
 510                 efi_systab.con_out_handle = systab64->con_out_handle;
 511                 tmp |= systab64->con_out_handle;
 512                 efi_systab.con_out = systab64->con_out;
 513                 tmp |= systab64->con_out;
 514                 efi_systab.stderr_handle = systab64->stderr_handle;
 515                 tmp |= systab64->stderr_handle;
 516                 efi_systab.stderr = systab64->stderr;
 517                 tmp |= systab64->stderr;
 518                 efi_systab.runtime = data ?
 519                                      (void *)(unsigned long)data->runtime :
 520                                      (void *)(unsigned long)systab64->runtime;
 521                 tmp |= data ? data->runtime : systab64->runtime;
 522                 efi_systab.boottime = (void *)(unsigned long)systab64->boottime;
 523                 tmp |= systab64->boottime;
 524                 efi_systab.nr_tables = systab64->nr_tables;
 525                 efi_systab.tables = data ? (unsigned long)data->tables :
 526                                            systab64->tables;
 527                 tmp |= data ? data->tables : systab64->tables;
 528
 529                 early_iounmap(systab64, sizeof(*systab64));
 530                 if (data)
 531                         early_iounmap(data, sizeof(*data));
 532 #ifdef CONFIG_X86_32
 533                 if (tmp >> 32) {
 534                         pr_err("EFI data located above 4GB, disabling EFI.\n");
 535                         return -EINVAL;
 536                 }
 537 #endif
 538         } else {
 539                 efi_system_table_32_t *systab32;
 540
 541                 systab32 = early_ioremap((unsigned long)phys,
 542                                          sizeof(*systab32));
 543                 if (systab32 == NULL) {
 544                         pr_err("Couldn't map the system table!\n");
 545                         return -ENOMEM;
 546                 }
 547
 548                 efi_systab.hdr = systab32->hdr;
 549                 efi_systab.fw_vendor = systab32->fw_vendor;
 550                 efi_systab.fw_revision = systab32->fw_revision;
 551                 efi_systab.con_in_handle = systab32->con_in_handle;
 552                 efi_systab.con_in = systab32->con_in;
 553                 efi_systab.con_out_handle = systab32->con_out_handle;
 554                 efi_systab.con_out = systab32->con_out;
 555                 efi_systab.stderr_handle = systab32->stderr_handle;
 556                 efi_systab.stderr = systab32->stderr;
 557                 efi_systab.runtime = (void *)(unsigned long)systab32->runtime;
 558                 efi_systab.boottime = (void *)(unsigned long)systab32->boottime;
 559                 efi_systab.nr_tables = systab32->nr_tables;
 560                 efi_systab.tables = systab32->tables;
 561
 562                 early_iounmap(systab32, sizeof(*systab32));
 563         }
 564
 565         efi.systab = &efi_systab;
 566
 567         /*
 568          * Verify the EFI Table
 569          */
 570         if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE) {
 571                 pr_err("System table signature incorrect!\n");
 572                 return -EINVAL;
 573         }
 574         if ((efi.systab->hdr.revision >> 16) == 0)
 575                 pr_err("Warning: System table version "
 576                        "%d.%02d, expected 1.00 or greater!\n",
 577                        efi.systab->hdr.revision >> 16,
 578                        efi.systab->hdr.revision & 0xffff);
 579
 580         set_bit(EFI_SYSTEM_TABLES, &efi.flags);
 581
 582         return 0;
 583 }
 584
 585 static int __init efi_runtime_init(void)
 586 {
 587         efi_runtime_services_t *runtime;
 588
 589         /*
 590          * Check out the runtime services table. We need to map
 591          * the runtime services table so that we can grab the physical
 592          * address of several of the EFI runtime functions, needed to
 593          * set the firmware into virtual mode.
 594          */
 595         runtime = early_ioremap((unsigned long)efi.systab->runtime,
 596                                 sizeof(efi_runtime_services_t));
 597         if (!runtime) {
 598                 pr_err("Could not map the runtime service table!\n");
 599                 return -ENOMEM;
 600         }
 601         /*
 602          * We will only need *early* access to the following
 603          * two EFI runtime services before set_virtual_address_map
 604          * is invoked.
 605          */
 606         efi_phys.get_time = (efi_get_time_t *)runtime->get_time;
 607         efi_phys.set_virtual_address_map =
 608                 (efi_set_virtual_address_map_t *)
 609                 runtime->set_virtual_address_map;
 610         /*
 611          * Make efi_get_time can be called before entering
 612          * virtual mode.
 613          */
 614         efi.get_time = phys_efi_get_time;
 615         early_iounmap(runtime, sizeof(efi_runtime_services_t));
 616
 617         set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
 618
 619         return 0;
 620 }
 621
 622 static int __init efi_memmap_init(void)
 623 {
 624         /* Map the EFI memory map */
 625         memmap.map = early_ioremap((unsigned long)memmap.phys_map,
 626                                    memmap.nr_map * memmap.desc_size);
 627         if (memmap.map == NULL) {
 628                 pr_err("Could not map the memory map!\n");
 629                 return -ENOMEM;
 630         }
 631         memmap.map_end = memmap.map + (memmap.nr_map * memmap.desc_size);
 632
 633         if (add_efi_memmap)
 634                 do_add_efi_memmap();
 635
 636         set_bit(EFI_MEMMAP, &efi.flags);
 637
 638         return 0;
 639 }
 640
 641 /*
 642  * A number of config table entries get remapped to virtual addresses
 643  * after entering EFI virtual mode. However, the kexec kernel requires
 644  * their physical addresses therefore we pass them via setup_data and
 645  * correct those entries to their respective physical addresses here.
 646  *
 647  * Currently only handles smbios which is necessary for some firmware
 648  * implementation.
 649  */
 650 static int __init efi_reuse_config(u64 tables, int nr_tables)
 651 {
 652         int i, sz, ret = 0;
 653         void *p, *tablep;
 654         struct efi_setup_data *data;
 655
 656         if (!efi_setup)
 657                 return 0;
 658
 659         if (!efi_enabled(EFI_64BIT))
 660                 return 0;
 661
 662         data = early_memremap(efi_setup, sizeof(*data));
 663         if (!data) {
 664                 ret = -ENOMEM;
 665                 goto out;
 666         }
 667
 668         if (!data->smbios)
 669                 goto out_memremap;
 670
 671         sz = sizeof(efi_config_table_64_t);
 672
 673         p = tablep = early_memremap(tables, nr_tables * sz);
 674         if (!p) {
 675                 pr_err("Could not map Configuration table!\n");
 676                 ret = -ENOMEM;
 677                 goto out_memremap;
 678         }
 679
 680         for (i = 0; i < efi.systab->nr_tables; i++) {
 681                 efi_guid_t guid;
 682
 683                 guid = ((efi_config_table_64_t *)p)->guid;
 684
 685                 if (!efi_guidcmp(guid, SMBIOS_TABLE_GUID))
 686                         ((efi_config_table_64_t *)p)->table = data->smbios;
 687                 p += sz;
 688         }
 689         early_iounmap(tablep, nr_tables * sz);
 690
 691 out_memremap:
 692         early_iounmap(data, sizeof(*data));
 693 out:
 694         return ret;
 695 }
 696
 697 void __init efi_init(void)
 698 {
 699         efi_char16_t *c16;
 700         char vendor[100] = "unknown";
 701         int i = 0;
 702         void *tmp;
 703
 704 #ifdef CONFIG_X86_32
 705         if (boot_params.efi_info.efi_systab_hi ||
 706             boot_params.efi_info.efi_memmap_hi) {
 707                 pr_info("Table located above 4GB, disabling EFI.\n");
 708                 return;
 709         }
 710         efi_phys.systab = (efi_system_table_t *)boot_params.efi_info.efi_systab;
 711 #else
 712         efi_phys.systab = (efi_system_table_t *)
 713                           (boot_params.efi_info.efi_systab |
 714                           ((__u64)boot_params.efi_info.efi_systab_hi<<32));
 715 #endif
 716
 717         if (efi_systab_init(efi_phys.systab))
 718                 return;
 719
 720         set_bit(EFI_SYSTEM_TABLES, &efi.flags);
 721
 722         efi.config_table = (unsigned long)efi.systab->tables;
 723         efi.fw_vendor    = (unsigned long)efi.systab->fw_vendor;
 724         efi.runtime      = (unsigned long)efi.systab->runtime;
 725
 726         /*
 727          * Show what we know for posterity
 728          */
 729         c16 = tmp = early_ioremap(efi.systab->fw_vendor, 2);
 730         if (c16) {
 731                 for (i = 0; i < sizeof(vendor) - 1 && *c16; ++i)
 732                         vendor[i] = *c16++;
 733                 vendor[i] = '\0';
 734         } else
 735                 pr_err("Could not map the firmware vendor!\n");
 736         early_iounmap(tmp, 2);
 737
 738         pr_info("EFI v%u.%.02u by %s\n",
 739                 efi.systab->hdr.revision >> 16,
 740                 efi.systab->hdr.revision & 0xffff, vendor);
 741
 742         if (efi_reuse_config(efi.systab->tables, efi.systab->nr_tables))
 743                 return;
 744
 745         if (efi_config_init(arch_tables))
 746                 return;
 747
 748         /*
 749          * Note: We currently don't support runtime services on an EFI
 750          * that doesn't match the kernel 32/64-bit mode.
 751          */
 752
 753         if (!efi_is_native())
 754                 pr_info("No EFI runtime due to 32/64-bit mismatch with kernel\n");
 755         else {
 756                 if (disable_runtime || efi_runtime_init())
 757                         return;
 758         }
 759         if (efi_memmap_init())
 760                 return;
 761
 762         set_bit(EFI_MEMMAP, &efi.flags);
 763
 764         print_efi_memmap();
 765 }
 766
 767 void __init efi_late_init(void)
 768 {
 769         efi_bgrt_init();
 770 }
 771
 772 void __init efi_set_executable(efi_memory_desc_t *md, bool executable)
 773 {
 774         u64 addr, npages;
 775
 776         addr = md->virt_addr;
 777         npages = md->num_pages;
 778
 779         memrange_efi_to_native(&addr, &npages);
 780
 781         if (executable)
 782                 set_memory_x(addr, npages);
 783         else
 784                 set_memory_nx(addr, npages);
 785 }
 786
 787 void __init runtime_code_page_mkexec(void)
 788 {
 789         efi_memory_desc_t *md;
 790         void *p;
 791
 792         /* Make EFI runtime service code area executable */
 793         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 794                 md = p;
 795
 796                 if (md->type != EFI_RUNTIME_SERVICES_CODE)
 797                         continue;
 798
 799                 efi_set_executable(md, true);
 800         }
 801 }
 802
 803 void efi_memory_uc(u64 addr, unsigned long size)
 804 {
 805         unsigned long page_shift = 1UL << EFI_PAGE_SHIFT;
 806         u64 npages;
 807
 808         npages = round_up(size, page_shift) / page_shift;
 809         memrange_efi_to_native(&addr, &npages);
 810         set_memory_uc(addr, npages);
 811 }
 812
 813 void __init old_map_region(efi_memory_desc_t *md)
 814 {
 815         u64 start_pfn, end_pfn, end;
 816         unsigned long size;
 817         void *va;
 818
 819         start_pfn = PFN_DOWN(md->phys_addr);
 820         size      = md->num_pages << PAGE_SHIFT;
 821         end       = md->phys_addr + size;
 822         end_pfn   = PFN_UP(end);
 823
 824         if (pfn_range_is_mapped(start_pfn, end_pfn)) {
 825                 va = __va(md->phys_addr);
 826
 827                 if (!(md->attribute & EFI_MEMORY_WB))
 828                         efi_memory_uc((u64)(unsigned long)va, size);
 829         } else
 830                 va = efi_ioremap(md->phys_addr, size,
 831                                  md->type, md->attribute);
 832
 833         md->virt_addr = (u64) (unsigned long) va;
 834         if (!va)
 835                 pr_err("ioremap of 0x%llX failed!\n",
 836                        (unsigned long long)md->phys_addr);
 837 }
 838
 839 /* Merge contiguous regions of the same type and attribute */
 840 static void __init efi_merge_regions(void)
 841 {
 842         void *p;
 843         efi_memory_desc_t *md, *prev_md = NULL;
 844
 845         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 846                 u64 prev_size;
 847                 md = p;
 848
 849                 if (!prev_md) {
 850                         prev_md = md;
 851                         continue;
 852                 }
 853
 854                 if (prev_md->type != md->type ||
 855                     prev_md->attribute != md->attribute) {
 856                         prev_md = md;
 857                         continue;
 858                 }
 859
 860                 prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
 861
 862                 if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
 863                         prev_md->num_pages += md->num_pages;
 864                         md->type = EFI_RESERVED_TYPE;
 865                         md->attribute = 0;
 866                         continue;
 867                 }
 868                 prev_md = md;
 869         }
 870 }
 871
 872 static void __init get_systab_virt_addr(efi_memory_desc_t *md)
 873 {
 874         unsigned long size;
 875         u64 end, systab;
 876
 877         size = md->num_pages << EFI_PAGE_SHIFT;
 878         end = md->phys_addr + size;
 879         systab = (u64)(unsigned long)efi_phys.systab;
 880         if (md->phys_addr <= systab && systab < end) {
 881                 systab += md->virt_addr - md->phys_addr;
 882                 efi.systab = (efi_system_table_t *)(unsigned long)systab;
 883         }
 884 }
 885
 886 static int __init save_runtime_map(void)
 887 {
 888         efi_memory_desc_t *md;
 889         void *tmp, *p, *q = NULL;
 890         int count = 0;
 891
 892         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 893                 md = p;
 894
 895                 if (!(md->attribute & EFI_MEMORY_RUNTIME) ||
 896                     (md->type == EFI_BOOT_SERVICES_CODE) ||
 897                     (md->type == EFI_BOOT_SERVICES_DATA))
 898                         continue;
 899                 tmp = krealloc(q, (count + 1) * memmap.desc_size, GFP_KERNEL);
 900                 if (!tmp)
 901                         goto out;
 902                 q = tmp;
 903
 904                 memcpy(q + count * memmap.desc_size, md, memmap.desc_size);
 905                 count++;
 906         }
 907
 908         efi_runtime_map_setup(q, count, memmap.desc_size);
 909
 910         return 0;
 911 out:
 912         kfree(q);
 913         return -ENOMEM;
 914 }
 915
 916 /*
 917  * Map efi regions which were passed via setup_data. The virt_addr is a fixed
 918  * addr which was used in first kernel of a kexec boot.
 919  */
 920 static void __init efi_map_regions_fixed(void)
 921 {
 922         void *p;
 923         efi_memory_desc_t *md;
 924
 925         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 926                 md = p;
 927                 efi_map_region_fixed(md); /* FIXME: add error handling */
 928                 get_systab_virt_addr(md);
 929         }
 930
 931 }
 932
 933 /*
 934  * Map efi memory ranges for runtime serivce and update new_memmap with virtual
 935  * addresses.
 936  */
 937 static void * __init efi_map_regions(int *count)
 938 {
 939         efi_memory_desc_t *md;
 940         void *p, *tmp, *new_memmap = NULL;
 941
 942         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 943                 md = p;
 944                 if (!(md->attribute & EFI_MEMORY_RUNTIME)) {
 945 #ifdef CONFIG_X86_64
 946                         if (md->type != EFI_BOOT_SERVICES_CODE &&
 947                             md->type != EFI_BOOT_SERVICES_DATA)
 948 #endif
 949                                 continue;
 950                 }
 951
 952                 efi_map_region(md);
 953                 get_systab_virt_addr(md);
 954
 955                 tmp = krealloc(new_memmap, (*count + 1) * memmap.desc_size,
 956                                GFP_KERNEL);
 957                 if (!tmp)
 958                         goto out;
 959                 new_memmap = tmp;
 960                 memcpy(new_memmap + (*count * memmap.desc_size), md,
 961                        memmap.desc_size);
 962                 (*count)++;
 963         }
 964
 965         return new_memmap;
 966 out:
 967         kfree(new_memmap);
 968         return NULL;
 969 }
 970
 971 /*
 972  * This function will switch the EFI runtime services to virtual mode.
 973  * Essentially, we look through the EFI memmap and map every region that
 974  * has the runtime attribute bit set in its memory descriptor into the
 975  * ->trampoline_pgd page table using a top-down VA allocation scheme.
 976  *
 977  * The old method which used to update that memory descriptor with the
 978  * virtual address obtained from ioremap() is still supported when the
 979  * kernel is booted with efi=old_map on its command line. Same old
 980  * method enabled the runtime services to be called without having to
 981  * thunk back into physical mode for every invocation.
 982  *
 983  * The new method does a pagetable switch in a preemption-safe manner
 984  * so that we're in a different address space when calling a runtime
 985  * function. For function arguments passing we do copy the PGDs of the
 986  * kernel page table into ->trampoline_pgd prior to each call.
 987  *
 988  * Specially for kexec boot, efi runtime maps in previous kernel should
 989  * be passed in via setup_data. In that case runtime ranges will be mapped
 990  * to the same virtual addresses as the first kernel.
 991  */
 992 void __init efi_enter_virtual_mode(void)
 993 {
 994         efi_status_t status;
 995         void *new_memmap = NULL;
 996         int err, count = 0;
 997
 998         efi.systab = NULL;
 999
1000         /*
1001          * We don't do virtual mode, since we don't do runtime services, on
1002          * non-native EFI
1003          */
1004         if (!efi_is_native()) {
1005                 efi_unmap_memmap();
1006                 return;
1007         }
1008
1009         if (efi_setup) {
1010                 efi_map_regions_fixed();
1011         } else {
1012                 efi_merge_regions();
1013                 new_memmap = efi_map_regions(&count);
1014                 if (!new_memmap) {
1015                         pr_err("Error reallocating memory, EFI runtime non-functional!\n");
1016                         return;
1017                 }
1018         }
1019
1020         err = save_runtime_map();
1021         if (err)
1022                 pr_err("Error saving runtime map, efi runtime on kexec non-functional!!\n");
1023
1024         BUG_ON(!efi.systab);
1025
1026         efi_setup_page_tables();
1027         efi_sync_low_kernel_mappings();
1028
1029         if (!efi_setup) {
1030                 status = phys_efi_set_virtual_address_map(
1031                         memmap.desc_size * count,
1032                         memmap.desc_size,
1033                         memmap.desc_version,
1034                         (efi_memory_desc_t *)__pa(new_memmap));
1035
1036                 if (status != EFI_SUCCESS) {
1037                         pr_alert("Unable to switch EFI into virtual mode (status=%lx)!\n",
1038                                  status);
1039                         panic("EFI call to SetVirtualAddressMap() failed!");
1040                 }
1041         }
1042
1043         /*
1044          * Now that EFI is in virtual mode, update the function
1045          * pointers in the runtime service table to the new virtual addresses.
1046          *
1047          * Call EFI services through wrapper functions.
1048          */
1049         efi.runtime_version = efi_systab.hdr.revision;
1050         efi.get_time = virt_efi_get_time;
1051         efi.set_time = virt_efi_set_time;
1052         efi.get_wakeup_time = virt_efi_get_wakeup_time;
1053         efi.set_wakeup_time = virt_efi_set_wakeup_time;
1054         efi.get_variable = virt_efi_get_variable;
1055         efi.get_next_variable = virt_efi_get_next_variable;
1056         efi.set_variable = virt_efi_set_variable;
1057         efi.get_next_high_mono_count = virt_efi_get_next_high_mono_count;
1058         efi.reset_system = virt_efi_reset_system;
1059         efi.set_virtual_address_map = NULL;
1060         efi.query_variable_info = virt_efi_query_variable_info;
1061         efi.update_capsule = virt_efi_update_capsule;
1062         efi.query_capsule_caps = virt_efi_query_capsule_caps;
1063
1064         efi_runtime_mkexec();
1065
1066         kfree(new_memmap);
1067
1068         /* clean DUMMY object */
1069         efi.set_variable(efi_dummy_name, &EFI_DUMMY_GUID,
1070                          EFI_VARIABLE_NON_VOLATILE |
1071                          EFI_VARIABLE_BOOTSERVICE_ACCESS |
1072                          EFI_VARIABLE_RUNTIME_ACCESS,
1073                          0, NULL);
1074 }
1075
1076 /*
1077  * Convenience functions to obtain memory types and attributes
1078  */
1079 u32 efi_mem_type(unsigned long phys_addr)
1080 {
1081         efi_memory_desc_t *md;
1082         void *p;
1083
1084         if (!efi_enabled(EFI_MEMMAP))
1085                 return 0;
1086
1087         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
1088                 md = p;
1089                 if ((md->phys_addr <= phys_addr) &&
1090                     (phys_addr < (md->phys_addr +
1091                                   (md->num_pages << EFI_PAGE_SHIFT))))
1092                         return md->type;
1093         }
1094         return 0;
1095 }
1096
1097 u64 efi_mem_attributes(unsigned long phys_addr)
1098 {
1099         efi_memory_desc_t *md;
1100         void *p;
1101
1102         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
1103                 md = p;
1104                 if ((md->phys_addr <= phys_addr) &&
1105                     (phys_addr < (md->phys_addr +
1106                                   (md->num_pages << EFI_PAGE_SHIFT))))
1107                         return md->attribute;
1108         }
1109         return 0;
1110 }
1111
1112 /*
1113  * Some firmware implementations refuse to boot if there's insufficient space
1114  * in the variable store. Ensure that we never use more than a safe limit.
1115  *
1116  * Return EFI_SUCCESS if it is safe to write 'size' bytes to the variable
1117  * store.
1118  */
1119 efi_status_t efi_query_variable_store(u32 attributes, unsigned long size)
1120 {
1121         efi_status_t status;
1122         u64 storage_size, remaining_size, max_size;
1123
1124         if (!(attributes & EFI_VARIABLE_NON_VOLATILE))
1125                 return 0;
1126
1127         status = efi.query_variable_info(attributes, &storage_size,
1128                                          &remaining_size, &max_size);
1129         if (status != EFI_SUCCESS)
1130                 return status;
1131
1132         /*
1133          * We account for that by refusing the write if permitting it would
1134          * reduce the available space to under 5KB. This figure was provided by
1135          * Samsung, so should be safe.
1136          */
1137         if ((remaining_size - size < EFI_MIN_RESERVE) &&
1138                 !efi_no_storage_paranoia) {
1139
1140                 /*
1141                  * Triggering garbage collection may require that the firmware
1142                  * generate a real EFI_OUT_OF_RESOURCES error. We can force
1143                  * that by attempting to use more space than is available.
1144                  */
1145                 unsigned long dummy_size = remaining_size + 1024;
1146                 void *dummy = kzalloc(dummy_size, GFP_ATOMIC);
1147
1148                 if (!dummy)
1149                         return EFI_OUT_OF_RESOURCES;
1150
1151                 status = efi.set_variable(efi_dummy_name, &EFI_DUMMY_GUID,
1152                                           EFI_VARIABLE_NON_VOLATILE |
1153                                           EFI_VARIABLE_BOOTSERVICE_ACCESS |
1154                                           EFI_VARIABLE_RUNTIME_ACCESS,
1155                                           dummy_size, dummy);
1156
1157                 if (status == EFI_SUCCESS) {
1158                         /*
1159                          * This should have failed, so if it didn't make sure
1160                          * that we delete it...
1161                          */
1162                         efi.set_variable(efi_dummy_name, &EFI_DUMMY_GUID,
1163                                          EFI_VARIABLE_NON_VOLATILE |
1164                                          EFI_VARIABLE_BOOTSERVICE_ACCESS |
1165                                          EFI_VARIABLE_RUNTIME_ACCESS,
1166                                          0, dummy);
1167                 }
1168
1169                 kfree(dummy);
1170
1171                 /*
1172                  * The runtime code may now have triggered a garbage collection
1173                  * run, so check the variable info again
1174                  */
1175                 status = efi.query_variable_info(attributes, &storage_size,
1176                                                  &remaining_size, &max_size);
1177
1178                 if (status != EFI_SUCCESS)
1179                         return status;
1180
1181                 /*
1182                  * There still isn't enough room, so return an error
1183                  */
1184                 if (remaining_size - size < EFI_MIN_RESERVE)
1185                         return EFI_OUT_OF_RESOURCES;
1186         }
1187
1188         return EFI_SUCCESS;
1189 }
1190 EXPORT_SYMBOL_GPL(efi_query_variable_store);
1191
1192 static int __init parse_efi_cmdline(char *str)
1193 {
1194         if (*str == '=')
1195                 str++;
1196
1197         if (!strncmp(str, "old_map", 7))
1198                 set_bit(EFI_OLD_MEMMAP, &efi.flags);
1199
1200         return 0;
1201 }
1202 early_param("efi", parse_efi_cmdline);