arch/x86/platform/efi/efi.c

   1 /*
   2  * Common EFI (Extensible Firmware Interface) support functions
   3  * Based on Extensible Firmware Interface Specification version 1.0
   4  *
   5  * Copyright (C) 1999 VA Linux Systems
   6  * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
   7  * Copyright (C) 1999-2002 Hewlett-Packard Co.
   8  *      David Mosberger-Tang <davidm@hpl.hp.com>
   9  *      Stephane Eranian <eranian@hpl.hp.com>
  10  * Copyright (C) 2005-2008 Intel Co.
  11  *      Fenghua Yu <fenghua.yu@intel.com>
  12  *      Bibo Mao <bibo.mao@intel.com>
  13  *      Chandramouli Narayanan <mouli@linux.intel.com>
  14  *      Huang Ying <ying.huang@intel.com>
  15  * Copyright (C) 2013 SuSE Labs
  16  *      Borislav Petkov <bp@suse.de> - runtime services VA mapping
  17  *
  18  * Copied from efi_32.c to eliminate the duplicated code between EFI
  19  * 32/64 support code. --ying 2007-10-26
  20  *
  21  * All EFI Runtime Services are not implemented yet as EFI only
  22  * supports physical mode addressing on SoftSDV. This is to be fixed
  23  * in a future version.  --drummond 1999-07-20
  24  *
  25  * Implemented EFI runtime services and virtual mode calls.  --davidm
  26  *
  27  * Goutham Rao: <goutham.rao@intel.com>
  28  *      Skip non-WB memory and ignore empty memory ranges.
  29  */
  30
  31 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  32
  33 #include <linux/kernel.h>
  34 #include <linux/init.h>
  35 #include <linux/efi.h>
  36 #include <linux/efi-bgrt.h>
  37 #include <linux/export.h>
  38 #include <linux/bootmem.h>
  39 #include <linux/slab.h>
  40 #include <linux/memblock.h>
  41 #include <linux/spinlock.h>
  42 #include <linux/uaccess.h>
  43 #include <linux/time.h>
  44 #include <linux/io.h>
  45 #include <linux/reboot.h>
  46 #include <linux/bcd.h>
  47
  48 #include <asm/setup.h>
  49 #include <asm/efi.h>
  50 #include <asm/time.h>
  51 #include <asm/cacheflush.h>
  52 #include <asm/tlbflush.h>
  53 #include <asm/x86_init.h>
  54 #include <asm/rtc.h>
  55 #include <asm/uv/uv.h>
  56
  57 #define EFI_DEBUG
  58
  59 struct efi_memory_map memmap;
  60
  61 static struct efi efi_phys __initdata;
  62 static efi_system_table_t efi_systab __initdata;
  63
  64 static efi_config_table_type_t arch_tables[] __initdata = {
  65 #ifdef CONFIG_X86_UV
  66         {UV_SYSTEM_TABLE_GUID, "UVsystab", &efi.uv_systab},
  67 #endif
  68         {NULL_GUID, NULL, NULL},
  69 };
  70
  71 u64 efi_setup;          /* efi setup_data physical address */
  72
  73 static bool disable_runtime __initdata = false;
  74 static int __init setup_noefi(char *arg)
  75 {
  76         disable_runtime = true;
  77         return 0;
  78 }
  79 early_param("noefi", setup_noefi);
  80
  81 int add_efi_memmap;
  82 EXPORT_SYMBOL(add_efi_memmap);
  83
  84 static int __init setup_add_efi_memmap(char *arg)
  85 {
  86         add_efi_memmap = 1;
  87         return 0;
  88 }
  89 early_param("add_efi_memmap", setup_add_efi_memmap);
  90
  91 static efi_status_t virt_efi_get_time(efi_time_t *tm, efi_time_cap_t *tc)
  92 {
  93         unsigned long flags;
  94         efi_status_t status;
  95
  96         spin_lock_irqsave(&rtc_lock, flags);
  97         status = efi_call_virt(get_time, tm, tc);
  98         spin_unlock_irqrestore(&rtc_lock, flags);
  99         return status;
 100 }
 101
 102 static efi_status_t virt_efi_set_time(efi_time_t *tm)
 103 {
 104         unsigned long flags;
 105         efi_status_t status;
 106
 107         spin_lock_irqsave(&rtc_lock, flags);
 108         status = efi_call_virt(set_time, tm);
 109         spin_unlock_irqrestore(&rtc_lock, flags);
 110         return status;
 111 }
 112
 113 static efi_status_t virt_efi_get_wakeup_time(efi_bool_t *enabled,
 114                                              efi_bool_t *pending,
 115                                              efi_time_t *tm)
 116 {
 117         unsigned long flags;
 118         efi_status_t status;
 119
 120         spin_lock_irqsave(&rtc_lock, flags);
 121         status = efi_call_virt(get_wakeup_time, enabled, pending, tm);
 122         spin_unlock_irqrestore(&rtc_lock, flags);
 123         return status;
 124 }
 125
 126 static efi_status_t virt_efi_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm)
 127 {
 128         unsigned long flags;
 129         efi_status_t status;
 130
 131         spin_lock_irqsave(&rtc_lock, flags);
 132         status = efi_call_virt(set_wakeup_time, enabled, tm);
 133         spin_unlock_irqrestore(&rtc_lock, flags);
 134         return status;
 135 }
 136
 137 static efi_status_t virt_efi_get_variable(efi_char16_t *name,
 138                                           efi_guid_t *vendor,
 139                                           u32 *attr,
 140                                           unsigned long *data_size,
 141                                           void *data)
 142 {
 143         return efi_call_virt(get_variable,
 144                              name, vendor, attr,
 145                              data_size, data);
 146 }
 147
 148 static efi_status_t virt_efi_get_next_variable(unsigned long *name_size,
 149                                                efi_char16_t *name,
 150                                                efi_guid_t *vendor)
 151 {
 152         return efi_call_virt(get_next_variable,
 153                              name_size, name, vendor);
 154 }
 155
 156 static efi_status_t virt_efi_set_variable(efi_char16_t *name,
 157                                           efi_guid_t *vendor,
 158                                           u32 attr,
 159                                           unsigned long data_size,
 160                                           void *data)
 161 {
 162         return efi_call_virt(set_variable,
 163                              name, vendor, attr,
 164                              data_size, data);
 165 }
 166
 167 static efi_status_t virt_efi_query_variable_info(u32 attr,
 168                                                  u64 *storage_space,
 169                                                  u64 *remaining_space,
 170                                                  u64 *max_variable_size)
 171 {
 172         if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
 173                 return EFI_UNSUPPORTED;
 174
 175         return efi_call_virt(query_variable_info, attr, storage_space,
 176                              remaining_space, max_variable_size);
 177 }
 178
 179 static efi_status_t virt_efi_get_next_high_mono_count(u32 *count)
 180 {
 181         return efi_call_virt(get_next_high_mono_count, count);
 182 }
 183
 184 static void virt_efi_reset_system(int reset_type,
 185                                   efi_status_t status,
 186                                   unsigned long data_size,
 187                                   efi_char16_t *data)
 188 {
 189         __efi_call_virt(reset_system, reset_type, status,
 190                         data_size, data);
 191 }
 192
 193 static efi_status_t virt_efi_update_capsule(efi_capsule_header_t **capsules,
 194                                             unsigned long count,
 195                                             unsigned long sg_list)
 196 {
 197         if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
 198                 return EFI_UNSUPPORTED;
 199
 200         return efi_call_virt(update_capsule, capsules, count, sg_list);
 201 }
 202
 203 static efi_status_t virt_efi_query_capsule_caps(efi_capsule_header_t **capsules,
 204                                                 unsigned long count,
 205                                                 u64 *max_size,
 206                                                 int *reset_type)
 207 {
 208         if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
 209                 return EFI_UNSUPPORTED;
 210
 211         return efi_call_virt(query_capsule_caps, capsules, count, max_size,
 212                              reset_type);
 213 }
 214
 215 static efi_status_t __init phys_efi_set_virtual_address_map(
 216         unsigned long memory_map_size,
 217         unsigned long descriptor_size,
 218         u32 descriptor_version,
 219         efi_memory_desc_t *virtual_map)
 220 {
 221         efi_status_t status;
 222
 223         efi_call_phys_prelog();
 224         status = efi_call_phys(efi_phys.set_virtual_address_map,
 225                                memory_map_size, descriptor_size,
 226                                descriptor_version, virtual_map);
 227         efi_call_phys_epilog();
 228         return status;
 229 }
 230
 231 int efi_set_rtc_mmss(const struct timespec *now)
 232 {
 233         unsigned long nowtime = now->tv_sec;
 234         efi_status_t    status;
 235         efi_time_t      eft;
 236         efi_time_cap_t  cap;
 237         struct rtc_time tm;
 238
 239         status = efi.get_time(&eft, &cap);
 240         if (status != EFI_SUCCESS) {
 241                 pr_err("Oops: efitime: can't read time!\n");
 242                 return -1;
 243         }
 244
 245         rtc_time_to_tm(nowtime, &tm);
 246         if (!rtc_valid_tm(&tm)) {
 247                 eft.year = tm.tm_year + 1900;
 248                 eft.month = tm.tm_mon + 1;
 249                 eft.day = tm.tm_mday;
 250                 eft.minute = tm.tm_min;
 251                 eft.second = tm.tm_sec;
 252                 eft.nanosecond = 0;
 253         } else {
 254                 pr_err("%s: Invalid EFI RTC value: write of %lx to EFI RTC failed\n",
 255                        __func__, nowtime);
 256                 return -1;
 257         }
 258
 259         status = efi.set_time(&eft);
 260         if (status != EFI_SUCCESS) {
 261                 pr_err("Oops: efitime: can't write time!\n");
 262                 return -1;
 263         }
 264         return 0;
 265 }
 266
 267 void efi_get_time(struct timespec *now)
 268 {
 269         efi_status_t status;
 270         efi_time_t eft;
 271         efi_time_cap_t cap;
 272
 273         status = efi.get_time(&eft, &cap);
 274         if (status != EFI_SUCCESS)
 275                 pr_err("Oops: efitime: can't read time!\n");
 276
 277         now->tv_sec = mktime(eft.year, eft.month, eft.day, eft.hour,
 278                              eft.minute, eft.second);
 279         now->tv_nsec = 0;
 280 }
 281
 282 /*
 283  * Tell the kernel about the EFI memory map.  This might include
 284  * more than the max 128 entries that can fit in the e820 legacy
 285  * (zeropage) memory map.
 286  */
 287
 288 static void __init do_add_efi_memmap(void)
 289 {
 290         void *p;
 291
 292         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 293                 efi_memory_desc_t *md = p;
 294                 unsigned long long start = md->phys_addr;
 295                 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
 296                 int e820_type;
 297
 298                 switch (md->type) {
 299                 case EFI_LOADER_CODE:
 300                 case EFI_LOADER_DATA:
 301                 case EFI_BOOT_SERVICES_CODE:
 302                 case EFI_BOOT_SERVICES_DATA:
 303                 case EFI_CONVENTIONAL_MEMORY:
 304                         if (md->attribute & EFI_MEMORY_WB)
 305                                 e820_type = E820_RAM;
 306                         else
 307                                 e820_type = E820_RESERVED;
 308                         break;
 309                 case EFI_ACPI_RECLAIM_MEMORY:
 310                         e820_type = E820_ACPI;
 311                         break;
 312                 case EFI_ACPI_MEMORY_NVS:
 313                         e820_type = E820_NVS;
 314                         break;
 315                 case EFI_UNUSABLE_MEMORY:
 316                         e820_type = E820_UNUSABLE;
 317                         break;
 318                 default:
 319                         /*
 320                          * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
 321                          * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
 322                          * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
 323                          */
 324                         e820_type = E820_RESERVED;
 325                         break;
 326                 }
 327                 e820_add_region(start, size, e820_type);
 328         }
 329         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 330 }
 331
 332 int __init efi_memblock_x86_reserve_range(void)
 333 {
 334         struct efi_info *e = &boot_params.efi_info;
 335         unsigned long pmap;
 336
 337 #ifdef CONFIG_X86_32
 338         /* Can't handle data above 4GB at this time */
 339         if (e->efi_memmap_hi) {
 340                 pr_err("Memory map is above 4GB, disabling EFI.\n");
 341                 return -EINVAL;
 342         }
 343         pmap =  e->efi_memmap;
 344 #else
 345         pmap = (e->efi_memmap | ((__u64)e->efi_memmap_hi << 32));
 346 #endif
 347         memmap.phys_map         = (void *)pmap;
 348         memmap.nr_map           = e->efi_memmap_size /
 349                                   e->efi_memdesc_size;
 350         memmap.desc_size        = e->efi_memdesc_size;
 351         memmap.desc_version     = e->efi_memdesc_version;
 352
 353         memblock_reserve(pmap, memmap.nr_map * memmap.desc_size);
 354
 355         efi.memmap = &memmap;
 356
 357         return 0;
 358 }
 359
 360 static void __init print_efi_memmap(void)
 361 {
 362 #ifdef EFI_DEBUG
 363         efi_memory_desc_t *md;
 364         void *p;
 365         int i;
 366
 367         for (p = memmap.map, i = 0;
 368              p < memmap.map_end;
 369              p += memmap.desc_size, i++) {
 370                 md = p;
 371                 pr_info("mem%02u: type=%u, attr=0x%llx, range=[0x%016llx-0x%016llx) (%lluMB)\n",
 372                         i, md->type, md->attribute, md->phys_addr,
 373                         md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT),
 374                         (md->num_pages >> (20 - EFI_PAGE_SHIFT)));
 375         }
 376 #endif  /*  EFI_DEBUG  */
 377 }
 378
 379 void __init efi_unmap_memmap(void)
 380 {
 381         clear_bit(EFI_MEMMAP, &efi.flags);
 382         if (memmap.map) {
 383                 early_iounmap(memmap.map, memmap.nr_map * memmap.desc_size);
 384                 memmap.map = NULL;
 385         }
 386 }
 387
 388 static int __init efi_systab_init(void *phys)
 389 {
 390         if (efi_enabled(EFI_64BIT)) {
 391                 efi_system_table_64_t *systab64;
 392                 struct efi_setup_data *data = NULL;
 393                 u64 tmp = 0;
 394
 395                 if (efi_setup) {
 396                         data = early_memremap(efi_setup, sizeof(*data));
 397                         if (!data)
 398                                 return -ENOMEM;
 399                 }
 400                 systab64 = early_ioremap((unsigned long)phys,
 401                                          sizeof(*systab64));
 402                 if (systab64 == NULL) {
 403                         pr_err("Couldn't map the system table!\n");
 404                         if (data)
 405                                 early_iounmap(data, sizeof(*data));
 406                         return -ENOMEM;
 407                 }
 408
 409                 efi_systab.hdr = systab64->hdr;
 410                 efi_systab.fw_vendor = data ? (unsigned long)data->fw_vendor :
 411                                               systab64->fw_vendor;
 412                 tmp |= data ? data->fw_vendor : systab64->fw_vendor;
 413                 efi_systab.fw_revision = systab64->fw_revision;
 414                 efi_systab.con_in_handle = systab64->con_in_handle;
 415                 tmp |= systab64->con_in_handle;
 416                 efi_systab.con_in = systab64->con_in;
 417                 tmp |= systab64->con_in;
 418                 efi_systab.con_out_handle = systab64->con_out_handle;
 419                 tmp |= systab64->con_out_handle;
 420                 efi_systab.con_out = systab64->con_out;
 421                 tmp |= systab64->con_out;
 422                 efi_systab.stderr_handle = systab64->stderr_handle;
 423                 tmp |= systab64->stderr_handle;
 424                 efi_systab.stderr = systab64->stderr;
 425                 tmp |= systab64->stderr;
 426                 efi_systab.runtime = data ?
 427                                      (void *)(unsigned long)data->runtime :
 428                                      (void *)(unsigned long)systab64->runtime;
 429                 tmp |= data ? data->runtime : systab64->runtime;
 430                 efi_systab.boottime = (void *)(unsigned long)systab64->boottime;
 431                 tmp |= systab64->boottime;
 432                 efi_systab.nr_tables = systab64->nr_tables;
 433                 efi_systab.tables = data ? (unsigned long)data->tables :
 434                                            systab64->tables;
 435                 tmp |= data ? data->tables : systab64->tables;
 436
 437                 early_iounmap(systab64, sizeof(*systab64));
 438                 if (data)
 439                         early_iounmap(data, sizeof(*data));
 440 #ifdef CONFIG_X86_32
 441                 if (tmp >> 32) {
 442                         pr_err("EFI data located above 4GB, disabling EFI.\n");
 443                         return -EINVAL;
 444                 }
 445 #endif
 446         } else {
 447                 efi_system_table_32_t *systab32;
 448
 449                 systab32 = early_ioremap((unsigned long)phys,
 450                                          sizeof(*systab32));
 451                 if (systab32 == NULL) {
 452                         pr_err("Couldn't map the system table!\n");
 453                         return -ENOMEM;
 454                 }
 455
 456                 efi_systab.hdr = systab32->hdr;
 457                 efi_systab.fw_vendor = systab32->fw_vendor;
 458                 efi_systab.fw_revision = systab32->fw_revision;
 459                 efi_systab.con_in_handle = systab32->con_in_handle;
 460                 efi_systab.con_in = systab32->con_in;
 461                 efi_systab.con_out_handle = systab32->con_out_handle;
 462                 efi_systab.con_out = systab32->con_out;
 463                 efi_systab.stderr_handle = systab32->stderr_handle;
 464                 efi_systab.stderr = systab32->stderr;
 465                 efi_systab.runtime = (void *)(unsigned long)systab32->runtime;
 466                 efi_systab.boottime = (void *)(unsigned long)systab32->boottime;
 467                 efi_systab.nr_tables = systab32->nr_tables;
 468                 efi_systab.tables = systab32->tables;
 469
 470                 early_iounmap(systab32, sizeof(*systab32));
 471         }
 472
 473         efi.systab = &efi_systab;
 474
 475         /*
 476          * Verify the EFI Table
 477          */
 478         if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE) {
 479                 pr_err("System table signature incorrect!\n");
 480                 return -EINVAL;
 481         }
 482         if ((efi.systab->hdr.revision >> 16) == 0)
 483                 pr_err("Warning: System table version %d.%02d, expected 1.00 or greater!\n",
 484                        efi.systab->hdr.revision >> 16,
 485                        efi.systab->hdr.revision & 0xffff);
 486
 487         set_bit(EFI_SYSTEM_TABLES, &efi.flags);
 488
 489         return 0;
 490 }
 491
 492 static int __init efi_runtime_init32(void)
 493 {
 494         efi_runtime_services_32_t *runtime;
 495
 496         runtime = early_ioremap((unsigned long)efi.systab->runtime,
 497                         sizeof(efi_runtime_services_32_t));
 498         if (!runtime) {
 499                 pr_err("Could not map the runtime service table!\n");
 500                 return -ENOMEM;
 501         }
 502
 503         /*
 504          * We will only need *early* access to the following two
 505          * EFI runtime services before set_virtual_address_map
 506          * is invoked.
 507          */
 508         efi_phys.set_virtual_address_map =
 509                         (efi_set_virtual_address_map_t *)
 510                         (unsigned long)runtime->set_virtual_address_map;
 511         early_iounmap(runtime, sizeof(efi_runtime_services_32_t));
 512
 513         return 0;
 514 }
 515
 516 static int __init efi_runtime_init64(void)
 517 {
 518         efi_runtime_services_64_t *runtime;
 519
 520         runtime = early_ioremap((unsigned long)efi.systab->runtime,
 521                         sizeof(efi_runtime_services_64_t));
 522         if (!runtime) {
 523                 pr_err("Could not map the runtime service table!\n");
 524                 return -ENOMEM;
 525         }
 526
 527         /*
 528          * We will only need *early* access to the following two
 529          * EFI runtime services before set_virtual_address_map
 530          * is invoked.
 531          */
 532         efi_phys.set_virtual_address_map =
 533                         (efi_set_virtual_address_map_t *)
 534                         (unsigned long)runtime->set_virtual_address_map;
 535         early_iounmap(runtime, sizeof(efi_runtime_services_64_t));
 536
 537         return 0;
 538 }
 539
 540 static int __init efi_runtime_init(void)
 541 {
 542         int rv;
 543
 544         /*
 545          * Check out the runtime services table. We need to map
 546          * the runtime services table so that we can grab the physical
 547          * address of several of the EFI runtime functions, needed to
 548          * set the firmware into virtual mode.
 549          */
 550         if (efi_enabled(EFI_64BIT))
 551                 rv = efi_runtime_init64();
 552         else
 553                 rv = efi_runtime_init32();
 554
 555         if (rv)
 556                 return rv;
 557
 558         set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
 559
 560         return 0;
 561 }
 562
 563 static int __init efi_memmap_init(void)
 564 {
 565         /* Map the EFI memory map */
 566         memmap.map = early_ioremap((unsigned long)memmap.phys_map,
 567                                    memmap.nr_map * memmap.desc_size);
 568         if (memmap.map == NULL) {
 569                 pr_err("Could not map the memory map!\n");
 570                 return -ENOMEM;
 571         }
 572         memmap.map_end = memmap.map + (memmap.nr_map * memmap.desc_size);
 573
 574         if (add_efi_memmap)
 575                 do_add_efi_memmap();
 576
 577         set_bit(EFI_MEMMAP, &efi.flags);
 578
 579         return 0;
 580 }
 581
 582 void __init efi_init(void)
 583 {
 584         efi_char16_t *c16;
 585         char vendor[100] = "unknown";
 586         int i = 0;
 587         void *tmp;
 588
 589 #ifdef CONFIG_X86_32
 590         if (boot_params.efi_info.efi_systab_hi ||
 591             boot_params.efi_info.efi_memmap_hi) {
 592                 pr_info("Table located above 4GB, disabling EFI.\n");
 593                 return;
 594         }
 595         efi_phys.systab = (efi_system_table_t *)boot_params.efi_info.efi_systab;
 596 #else
 597         efi_phys.systab = (efi_system_table_t *)
 598                           (boot_params.efi_info.efi_systab |
 599                           ((__u64)boot_params.efi_info.efi_systab_hi<<32));
 600 #endif
 601
 602         if (efi_systab_init(efi_phys.systab))
 603                 return;
 604
 605         set_bit(EFI_SYSTEM_TABLES, &efi.flags);
 606
 607         efi.config_table = (unsigned long)efi.systab->tables;
 608         efi.fw_vendor    = (unsigned long)efi.systab->fw_vendor;
 609         efi.runtime      = (unsigned long)efi.systab->runtime;
 610
 611         /*
 612          * Show what we know for posterity
 613          */
 614         c16 = tmp = early_ioremap(efi.systab->fw_vendor, 2);
 615         if (c16) {
 616                 for (i = 0; i < sizeof(vendor) - 1 && *c16; ++i)
 617                         vendor[i] = *c16++;
 618                 vendor[i] = '\0';
 619         } else
 620                 pr_err("Could not map the firmware vendor!\n");
 621         early_iounmap(tmp, 2);
 622
 623         pr_info("EFI v%u.%.02u by %s\n",
 624                 efi.systab->hdr.revision >> 16,
 625                 efi.systab->hdr.revision & 0xffff, vendor);
 626
 627         if (efi_reuse_config(efi.systab->tables, efi.systab->nr_tables))
 628                 return;
 629
 630         if (efi_config_init(arch_tables))
 631                 return;
 632
 633         /*
 634          * Note: We currently don't support runtime services on an EFI
 635          * that doesn't match the kernel 32/64-bit mode.
 636          */
 637
 638         if (!efi_runtime_supported())
 639                 pr_info("No EFI runtime due to 32/64-bit mismatch with kernel\n");
 640         else {
 641                 if (disable_runtime || efi_runtime_init())
 642                         return;
 643         }
 644         if (efi_memmap_init())
 645                 return;
 646
 647         set_bit(EFI_MEMMAP, &efi.flags);
 648
 649         print_efi_memmap();
 650 }
 651
 652 void __init efi_late_init(void)
 653 {
 654         efi_bgrt_init();
 655 }
 656
 657 void __init efi_set_executable(efi_memory_desc_t *md, bool executable)
 658 {
 659         u64 addr, npages;
 660
 661         addr = md->virt_addr;
 662         npages = md->num_pages;
 663
 664         memrange_efi_to_native(&addr, &npages);
 665
 666         if (executable)
 667                 set_memory_x(addr, npages);
 668         else
 669                 set_memory_nx(addr, npages);
 670 }
 671
 672 void __init runtime_code_page_mkexec(void)
 673 {
 674         efi_memory_desc_t *md;
 675         void *p;
 676
 677         /* Make EFI runtime service code area executable */
 678         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 679                 md = p;
 680
 681                 if (md->type != EFI_RUNTIME_SERVICES_CODE)
 682                         continue;
 683
 684                 efi_set_executable(md, true);
 685         }
 686 }
 687
 688 void efi_memory_uc(u64 addr, unsigned long size)
 689 {
 690         unsigned long page_shift = 1UL << EFI_PAGE_SHIFT;
 691         u64 npages;
 692
 693         npages = round_up(size, page_shift) / page_shift;
 694         memrange_efi_to_native(&addr, &npages);
 695         set_memory_uc(addr, npages);
 696 }
 697
 698 void __init old_map_region(efi_memory_desc_t *md)
 699 {
 700         u64 start_pfn, end_pfn, end;
 701         unsigned long size;
 702         void *va;
 703
 704         start_pfn = PFN_DOWN(md->phys_addr);
 705         size      = md->num_pages << PAGE_SHIFT;
 706         end       = md->phys_addr + size;
 707         end_pfn   = PFN_UP(end);
 708
 709         if (pfn_range_is_mapped(start_pfn, end_pfn)) {
 710                 va = __va(md->phys_addr);
 711
 712                 if (!(md->attribute & EFI_MEMORY_WB))
 713                         efi_memory_uc((u64)(unsigned long)va, size);
 714         } else
 715                 va = efi_ioremap(md->phys_addr, size,
 716                                  md->type, md->attribute);
 717
 718         md->virt_addr = (u64) (unsigned long) va;
 719         if (!va)
 720                 pr_err("ioremap of 0x%llX failed!\n",
 721                        (unsigned long long)md->phys_addr);
 722 }
 723
 724 static void native_runtime_setup(void)
 725 {
 726         efi.get_time = virt_efi_get_time;
 727         efi.set_time = virt_efi_set_time;
 728         efi.get_wakeup_time = virt_efi_get_wakeup_time;
 729         efi.set_wakeup_time = virt_efi_set_wakeup_time;
 730         efi.get_variable = virt_efi_get_variable;
 731         efi.get_next_variable = virt_efi_get_next_variable;
 732         efi.set_variable = virt_efi_set_variable;
 733         efi.get_next_high_mono_count = virt_efi_get_next_high_mono_count;
 734         efi.reset_system = virt_efi_reset_system;
 735         efi.query_variable_info = virt_efi_query_variable_info;
 736         efi.update_capsule = virt_efi_update_capsule;
 737         efi.query_capsule_caps = virt_efi_query_capsule_caps;
 738 }
 739
 740 /* Merge contiguous regions of the same type and attribute */
 741 static void __init efi_merge_regions(void)
 742 {
 743         void *p;
 744         efi_memory_desc_t *md, *prev_md = NULL;
 745
 746         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 747                 u64 prev_size;
 748                 md = p;
 749
 750                 if (!prev_md) {
 751                         prev_md = md;
 752                         continue;
 753                 }
 754
 755                 if (prev_md->type != md->type ||
 756                     prev_md->attribute != md->attribute) {
 757                         prev_md = md;
 758                         continue;
 759                 }
 760
 761                 prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
 762
 763                 if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
 764                         prev_md->num_pages += md->num_pages;
 765                         md->type = EFI_RESERVED_TYPE;
 766                         md->attribute = 0;
 767                         continue;
 768                 }
 769                 prev_md = md;
 770         }
 771 }
 772
 773 static void __init get_systab_virt_addr(efi_memory_desc_t *md)
 774 {
 775         unsigned long size;
 776         u64 end, systab;
 777
 778         size = md->num_pages << EFI_PAGE_SHIFT;
 779         end = md->phys_addr + size;
 780         systab = (u64)(unsigned long)efi_phys.systab;
 781         if (md->phys_addr <= systab && systab < end) {
 782                 systab += md->virt_addr - md->phys_addr;
 783                 efi.systab = (efi_system_table_t *)(unsigned long)systab;
 784         }
 785 }
 786
 787 static void __init save_runtime_map(void)
 788 {
 789 #ifdef CONFIG_KEXEC
 790         efi_memory_desc_t *md;
 791         void *tmp, *p, *q = NULL;
 792         int count = 0;
 793
 794         if (efi_enabled(EFI_OLD_MEMMAP))
 795                 return;
 796
 797         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 798                 md = p;
 799
 800                 if (!(md->attribute & EFI_MEMORY_RUNTIME) ||
 801                     (md->type == EFI_BOOT_SERVICES_CODE) ||
 802                     (md->type == EFI_BOOT_SERVICES_DATA))
 803                         continue;
 804                 tmp = krealloc(q, (count + 1) * memmap.desc_size, GFP_KERNEL);
 805                 if (!tmp)
 806                         goto out;
 807                 q = tmp;
 808
 809                 memcpy(q + count * memmap.desc_size, md, memmap.desc_size);
 810                 count++;
 811         }
 812
 813         efi_runtime_map_setup(q, count, memmap.desc_size);
 814         return;
 815
 816 out:
 817         kfree(q);
 818         pr_err("Error saving runtime map, efi runtime on kexec non-functional!!\n");
 819 #endif
 820 }
 821
 822 static void *realloc_pages(void *old_memmap, int old_shift)
 823 {
 824         void *ret;
 825
 826         ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1);
 827         if (!ret)
 828                 goto out;
 829
 830         /*
 831          * A first-time allocation doesn't have anything to copy.
 832          */
 833         if (!old_memmap)
 834                 return ret;
 835
 836         memcpy(ret, old_memmap, PAGE_SIZE << old_shift);
 837
 838 out:
 839         free_pages((unsigned long)old_memmap, old_shift);
 840         return ret;
 841 }
 842
 843 /*
 844  * Map the efi memory ranges of the runtime services and update new_mmap with
 845  * virtual addresses.
 846  */
 847 static void * __init efi_map_regions(int *count, int *pg_shift)
 848 {
 849         void *p, *new_memmap = NULL;
 850         unsigned long left = 0;
 851         efi_memory_desc_t *md;
 852
 853         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 854                 md = p;
 855                 if (!(md->attribute & EFI_MEMORY_RUNTIME)) {
 856 #ifdef CONFIG_X86_64
 857                         if (md->type != EFI_BOOT_SERVICES_CODE &&
 858                             md->type != EFI_BOOT_SERVICES_DATA)
 859 #endif
 860                                 continue;
 861                 }
 862
 863                 efi_map_region(md);
 864                 get_systab_virt_addr(md);
 865
 866                 if (left < memmap.desc_size) {
 867                         new_memmap = realloc_pages(new_memmap, *pg_shift);
 868                         if (!new_memmap)
 869                                 return NULL;
 870
 871                         left += PAGE_SIZE << *pg_shift;
 872                         (*pg_shift)++;
 873                 }
 874
 875                 memcpy(new_memmap + (*count * memmap.desc_size), md,
 876                        memmap.desc_size);
 877
 878                 left -= memmap.desc_size;
 879                 (*count)++;
 880         }
 881
 882         return new_memmap;
 883 }
 884
 885 static void __init kexec_enter_virtual_mode(void)
 886 {
 887 #ifdef CONFIG_KEXEC
 888         efi_memory_desc_t *md;
 889         void *p;
 890
 891         efi.systab = NULL;
 892
 893         /*
 894          * We don't do virtual mode, since we don't do runtime services, on
 895          * non-native EFI
 896          */
 897         if (!efi_is_native()) {
 898                 efi_unmap_memmap();
 899                 return;
 900         }
 901
 902         /*
 903         * Map efi regions which were passed via setup_data. The virt_addr is a
 904         * fixed addr which was used in first kernel of a kexec boot.
 905         */
 906         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 907                 md = p;
 908                 efi_map_region_fixed(md); /* FIXME: add error handling */
 909                 get_systab_virt_addr(md);
 910         }
 911
 912         save_runtime_map();
 913
 914         BUG_ON(!efi.systab);
 915
 916         efi_sync_low_kernel_mappings();
 917
 918         /*
 919          * Now that EFI is in virtual mode, update the function
 920          * pointers in the runtime service table to the new virtual addresses.
 921          *
 922          * Call EFI services through wrapper functions.
 923          */
 924         efi.runtime_version = efi_systab.hdr.revision;
 925
 926         native_runtime_setup();
 927
 928         efi.set_virtual_address_map = NULL;
 929
 930         if (efi_enabled(EFI_OLD_MEMMAP) && (__supported_pte_mask & _PAGE_NX))
 931                 runtime_code_page_mkexec();
 932
 933         /* clean DUMMY object */
 934         efi_delete_dummy_variable();
 935 #endif
 936 }
 937
 938 /*
 939  * This function will switch the EFI runtime services to virtual mode.
 940  * Essentially, we look through the EFI memmap and map every region that
 941  * has the runtime attribute bit set in its memory descriptor into the
 942  * ->trampoline_pgd page table using a top-down VA allocation scheme.
 943  *
 944  * The old method which used to update that memory descriptor with the
 945  * virtual address obtained from ioremap() is still supported when the
 946  * kernel is booted with efi=old_map on its command line. Same old
 947  * method enabled the runtime services to be called without having to
 948  * thunk back into physical mode for every invocation.
 949  *
 950  * The new method does a pagetable switch in a preemption-safe manner
 951  * so that we're in a different address space when calling a runtime
 952  * function. For function arguments passing we do copy the PGDs of the
 953  * kernel page table into ->trampoline_pgd prior to each call.
 954  *
 955  * Specially for kexec boot, efi runtime maps in previous kernel should
 956  * be passed in via setup_data. In that case runtime ranges will be mapped
 957  * to the same virtual addresses as the first kernel, see
 958  * kexec_enter_virtual_mode().
 959  */
 960 static void __init __efi_enter_virtual_mode(void)
 961 {
 962         int count = 0, pg_shift = 0;
 963         void *new_memmap = NULL;
 964         efi_status_t status;
 965
 966         efi.systab = NULL;
 967
 968         efi_merge_regions();
 969         new_memmap = efi_map_regions(&count, &pg_shift);
 970         if (!new_memmap) {
 971                 pr_err("Error reallocating memory, EFI runtime non-functional!\n");
 972                 return;
 973         }
 974
 975         save_runtime_map();
 976
 977         BUG_ON(!efi.systab);
 978
 979         if (efi_setup_page_tables(__pa(new_memmap), 1 << pg_shift))
 980                 return;
 981
 982         efi_sync_low_kernel_mappings();
 983         efi_dump_pagetable();
 984
 985         if (efi_is_native()) {
 986                 status = phys_efi_set_virtual_address_map(
 987                                 memmap.desc_size * count,
 988                                 memmap.desc_size,
 989                                 memmap.desc_version,
 990                                 (efi_memory_desc_t *)__pa(new_memmap));
 991         } else {
 992                 status = efi_thunk_set_virtual_address_map(
 993                                 efi_phys.set_virtual_address_map,
 994                                 memmap.desc_size * count,
 995                                 memmap.desc_size,
 996                                 memmap.desc_version,
 997                                 (efi_memory_desc_t *)__pa(new_memmap));
 998         }
 999
1000         if (status != EFI_SUCCESS) {
1001                 pr_alert("Unable to switch EFI into virtual mode (status=%lx)!\n",
1002                          status);
1003                 panic("EFI call to SetVirtualAddressMap() failed!");
1004         }
1005
1006         /*
1007          * Now that EFI is in virtual mode, update the function
1008          * pointers in the runtime service table to the new virtual addresses.
1009          *
1010          * Call EFI services through wrapper functions.
1011          */
1012         efi.runtime_version = efi_systab.hdr.revision;
1013
1014         if (efi_is_native())
1015                 native_runtime_setup();
1016         else
1017                 efi_thunk_runtime_setup();
1018
1019         efi.set_virtual_address_map = NULL;
1020
1021         efi_runtime_mkexec();
1022
1023         /*
1024          * We mapped the descriptor array into the EFI pagetable above but we're
1025          * not unmapping it here. Here's why:
1026          *
1027          * We're copying select PGDs from the kernel page table to the EFI page
1028          * table and when we do so and make changes to those PGDs like unmapping
1029          * stuff from them, those changes appear in the kernel page table and we
1030          * go boom.
1031          *
1032          * From setup_real_mode():
1033          *
1034          * ...
1035          * trampoline_pgd[0] = init_level4_pgt[pgd_index(__PAGE_OFFSET)].pgd;
1036          *
1037          * In this particular case, our allocation is in PGD 0 of the EFI page
1038          * table but we've copied that PGD from PGD[272] of the EFI page table:
1039          *
1040          *      pgd_index(__PAGE_OFFSET = 0xffff880000000000) = 272
1041          *
1042          * where the direct memory mapping in kernel space is.
1043          *
1044          * new_memmap's VA comes from that direct mapping and thus clearing it,
1045          * it would get cleared in the kernel page table too.
1046          *
1047          * efi_cleanup_page_tables(__pa(new_memmap), 1 << pg_shift);
1048          */
1049         free_pages((unsigned long)new_memmap, pg_shift);
1050
1051         /* clean DUMMY object */
1052         efi_delete_dummy_variable();
1053 }
1054
1055 void __init efi_enter_virtual_mode(void)
1056 {
1057         if (efi_setup)
1058                 kexec_enter_virtual_mode();
1059         else
1060                 __efi_enter_virtual_mode();
1061 }
1062
1063 /*
1064  * Convenience functions to obtain memory types and attributes
1065  */
1066 u32 efi_mem_type(unsigned long phys_addr)
1067 {
1068         efi_memory_desc_t *md;
1069         void *p;
1070
1071         if (!efi_enabled(EFI_MEMMAP))
1072                 return 0;
1073
1074         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
1075                 md = p;
1076                 if ((md->phys_addr <= phys_addr) &&
1077                     (phys_addr < (md->phys_addr +
1078                                   (md->num_pages << EFI_PAGE_SHIFT))))
1079                         return md->type;
1080         }
1081         return 0;
1082 }
1083
1084 u64 efi_mem_attributes(unsigned long phys_addr)
1085 {
1086         efi_memory_desc_t *md;
1087         void *p;
1088
1089         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
1090                 md = p;
1091                 if ((md->phys_addr <= phys_addr) &&
1092                     (phys_addr < (md->phys_addr +
1093                                   (md->num_pages << EFI_PAGE_SHIFT))))
1094                         return md->attribute;
1095         }
1096         return 0;
1097 }
1098
1099 static int __init parse_efi_cmdline(char *str)
1100 {
1101         if (*str == '=')
1102                 str++;
1103
1104         if (!strncmp(str, "old_map", 7))
1105                 set_bit(EFI_OLD_MEMMAP, &efi.flags);
1106
1107         return 0;
1108 }
1109 early_param("efi", parse_efi_cmdline);