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 59struct efi_memory_map memmap; 60 61static struct efi efi_phys __initdata; 62static efi_system_table_t efi_systab __initdata; 63 64static 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 71u64 efi_setup; /* efi setup_data physical address */ 72 73static int add_efi_memmap __initdata; 74static int __init setup_add_efi_memmap(char *arg) 75{ 76 add_efi_memmap = 1; 77 return 0; 78} 79early_param("add_efi_memmap", setup_add_efi_memmap); 80 81static efi_status_t __init phys_efi_set_virtual_address_map( 82 unsigned long memory_map_size, 83 unsigned long descriptor_size, 84 u32 descriptor_version, 85 efi_memory_desc_t *virtual_map) 86{ 87 efi_status_t status; 88 89 efi_call_phys_prolog(); 90 status = efi_call_phys(efi_phys.set_virtual_address_map, 91 memory_map_size, descriptor_size, 92 descriptor_version, virtual_map); 93 efi_call_phys_epilog(); 94 return status; 95} 96 97void efi_get_time(struct timespec *now) 98{ 99 efi_status_t status; 100 efi_time_t eft; 101 efi_time_cap_t cap; 102 103 status = efi.get_time(&eft, &cap); 104 if (status != EFI_SUCCESS) 105 pr_err("Oops: efitime: can't read time!\n"); 106 107 now->tv_sec = mktime(eft.year, eft.month, eft.day, eft.hour, 108 eft.minute, eft.second); 109 now->tv_nsec = 0; 110} 111 112/* 113 * Tell the kernel about the EFI memory map. This might include 114 * more than the max 128 entries that can fit in the e820 legacy 115 * (zeropage) memory map. 116 */ 117 118static void __init do_add_efi_memmap(void) 119{ 120 void *p; 121 122 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) { 123 efi_memory_desc_t *md = p; 124 unsigned long long start = md->phys_addr; 125 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT; 126 int e820_type; 127 128 switch (md->type) { 129 case EFI_LOADER_CODE: 130 case EFI_LOADER_DATA: 131 case EFI_BOOT_SERVICES_CODE: 132 case EFI_BOOT_SERVICES_DATA: 133 case EFI_CONVENTIONAL_MEMORY: 134 if (md->attribute & EFI_MEMORY_WB) 135 e820_type = E820_RAM; 136 else 137 e820_type = E820_RESERVED; 138 break; 139 case EFI_ACPI_RECLAIM_MEMORY: 140 e820_type = E820_ACPI; 141 break; 142 case EFI_ACPI_MEMORY_NVS: 143 e820_type = E820_NVS; 144 break; 145 case EFI_UNUSABLE_MEMORY: 146 e820_type = E820_UNUSABLE; 147 break; 148 default: 149 /* 150 * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE 151 * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO 152 * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE 153 */ 154 e820_type = E820_RESERVED; 155 break; 156 } 157 e820_add_region(start, size, e820_type); 158 } 159 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 160} 161 162int __init efi_memblock_x86_reserve_range(void) 163{ 164 struct efi_info *e = &boot_params.efi_info; 165 unsigned long pmap; 166 167 if (efi_enabled(EFI_PARAVIRT)) 168 return 0; 169 170#ifdef CONFIG_X86_32 171 /* Can't handle data above 4GB at this time */ 172 if (e->efi_memmap_hi) { 173 pr_err("Memory map is above 4GB, disabling EFI.\n"); 174 return -EINVAL; 175 } 176 pmap = e->efi_memmap; 177#else 178 pmap = (e->efi_memmap | ((__u64)e->efi_memmap_hi << 32)); 179#endif 180 memmap.phys_map = (void *)pmap; 181 memmap.nr_map = e->efi_memmap_size / 182 e->efi_memdesc_size; 183 memmap.desc_size = e->efi_memdesc_size; 184 memmap.desc_version = e->efi_memdesc_version; 185 186 memblock_reserve(pmap, memmap.nr_map * memmap.desc_size); 187 188 efi.memmap = &memmap; 189 190 return 0; 191} 192 193static void __init print_efi_memmap(void) 194{ 195#ifdef EFI_DEBUG 196 efi_memory_desc_t *md; 197 void *p; 198 int i; 199 200 for (p = memmap.map, i = 0; 201 p < memmap.map_end; 202 p += memmap.desc_size, i++) { 203 char buf[64]; 204 205 md = p; 206 pr_info("mem%02u: %s range=[0x%016llx-0x%016llx) (%lluMB)\n", 207 i, efi_md_typeattr_format(buf, sizeof(buf), md), 208 md->phys_addr, 209 md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT), 210 (md->num_pages >> (20 - EFI_PAGE_SHIFT))); 211 } 212#endif /* EFI_DEBUG */ 213} 214 215void __init efi_unmap_memmap(void) 216{ 217 clear_bit(EFI_MEMMAP, &efi.flags); 218 if (memmap.map) { 219 early_memunmap(memmap.map, memmap.nr_map * memmap.desc_size); 220 memmap.map = NULL; 221 } 222} 223 224static int __init efi_systab_init(void *phys) 225{ 226 if (efi_enabled(EFI_64BIT)) { 227 efi_system_table_64_t *systab64; 228 struct efi_setup_data *data = NULL; 229 u64 tmp = 0; 230 231 if (efi_setup) { 232 data = early_memremap(efi_setup, sizeof(*data)); 233 if (!data) 234 return -ENOMEM; 235 } 236 systab64 = early_memremap((unsigned long)phys, 237 sizeof(*systab64)); 238 if (systab64 == NULL) { 239 pr_err("Couldn't map the system table!\n"); 240 if (data) 241 early_memunmap(data, sizeof(*data)); 242 return -ENOMEM; 243 } 244 245 efi_systab.hdr = systab64->hdr; 246 efi_systab.fw_vendor = data ? (unsigned long)data->fw_vendor : 247 systab64->fw_vendor; 248 tmp |= data ? data->fw_vendor : systab64->fw_vendor; 249 efi_systab.fw_revision = systab64->fw_revision; 250 efi_systab.con_in_handle = systab64->con_in_handle; 251 tmp |= systab64->con_in_handle; 252 efi_systab.con_in = systab64->con_in; 253 tmp |= systab64->con_in; 254 efi_systab.con_out_handle = systab64->con_out_handle; 255 tmp |= systab64->con_out_handle; 256 efi_systab.con_out = systab64->con_out; 257 tmp |= systab64->con_out; 258 efi_systab.stderr_handle = systab64->stderr_handle; 259 tmp |= systab64->stderr_handle; 260 efi_systab.stderr = systab64->stderr; 261 tmp |= systab64->stderr; 262 efi_systab.runtime = data ? 263 (void *)(unsigned long)data->runtime : 264 (void *)(unsigned long)systab64->runtime; 265 tmp |= data ? data->runtime : systab64->runtime; 266 efi_systab.boottime = (void *)(unsigned long)systab64->boottime; 267 tmp |= systab64->boottime; 268 efi_systab.nr_tables = systab64->nr_tables; 269 efi_systab.tables = data ? (unsigned long)data->tables : 270 systab64->tables; 271 tmp |= data ? data->tables : systab64->tables; 272 273 early_memunmap(systab64, sizeof(*systab64)); 274 if (data) 275 early_memunmap(data, sizeof(*data)); 276#ifdef CONFIG_X86_32 277 if (tmp >> 32) { 278 pr_err("EFI data located above 4GB, disabling EFI.\n"); 279 return -EINVAL; 280 } 281#endif 282 } else { 283 efi_system_table_32_t *systab32; 284 285 systab32 = early_memremap((unsigned long)phys, 286 sizeof(*systab32)); 287 if (systab32 == NULL) { 288 pr_err("Couldn't map the system table!\n"); 289 return -ENOMEM; 290 } 291 292 efi_systab.hdr = systab32->hdr; 293 efi_systab.fw_vendor = systab32->fw_vendor; 294 efi_systab.fw_revision = systab32->fw_revision; 295 efi_systab.con_in_handle = systab32->con_in_handle; 296 efi_systab.con_in = systab32->con_in; 297 efi_systab.con_out_handle = systab32->con_out_handle; 298 efi_systab.con_out = systab32->con_out; 299 efi_systab.stderr_handle = systab32->stderr_handle; 300 efi_systab.stderr = systab32->stderr; 301 efi_systab.runtime = (void *)(unsigned long)systab32->runtime; 302 efi_systab.boottime = (void *)(unsigned long)systab32->boottime; 303 efi_systab.nr_tables = systab32->nr_tables; 304 efi_systab.tables = systab32->tables; 305 306 early_memunmap(systab32, sizeof(*systab32)); 307 } 308 309 efi.systab = &efi_systab; 310 311 /* 312 * Verify the EFI Table 313 */ 314 if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE) { 315 pr_err("System table signature incorrect!\n"); 316 return -EINVAL; 317 } 318 if ((efi.systab->hdr.revision >> 16) == 0) 319 pr_err("Warning: System table version %d.%02d, expected 1.00 or greater!\n", 320 efi.systab->hdr.revision >> 16, 321 efi.systab->hdr.revision & 0xffff); 322 323 set_bit(EFI_SYSTEM_TABLES, &efi.flags); 324 325 return 0; 326} 327 328static int __init efi_runtime_init32(void) 329{ 330 efi_runtime_services_32_t *runtime; 331 332 runtime = early_memremap((unsigned long)efi.systab->runtime, 333 sizeof(efi_runtime_services_32_t)); 334 if (!runtime) { 335 pr_err("Could not map the runtime service table!\n"); 336 return -ENOMEM; 337 } 338 339 /* 340 * We will only need *early* access to the SetVirtualAddressMap 341 * EFI runtime service. All other runtime services will be called 342 * via the virtual mapping. 343 */ 344 efi_phys.set_virtual_address_map = 345 (efi_set_virtual_address_map_t *) 346 (unsigned long)runtime->set_virtual_address_map; 347 early_memunmap(runtime, sizeof(efi_runtime_services_32_t)); 348 349 return 0; 350} 351 352static int __init efi_runtime_init64(void) 353{ 354 efi_runtime_services_64_t *runtime; 355 356 runtime = early_memremap((unsigned long)efi.systab->runtime, 357 sizeof(efi_runtime_services_64_t)); 358 if (!runtime) { 359 pr_err("Could not map the runtime service table!\n"); 360 return -ENOMEM; 361 } 362 363 /* 364 * We will only need *early* access to the SetVirtualAddressMap 365 * EFI runtime service. All other runtime services will be called 366 * via the virtual mapping. 367 */ 368 efi_phys.set_virtual_address_map = 369 (efi_set_virtual_address_map_t *) 370 (unsigned long)runtime->set_virtual_address_map; 371 early_memunmap(runtime, sizeof(efi_runtime_services_64_t)); 372 373 return 0; 374} 375 376static int __init efi_runtime_init(void) 377{ 378 int rv; 379 380 /* 381 * Check out the runtime services table. We need to map 382 * the runtime services table so that we can grab the physical 383 * address of several of the EFI runtime functions, needed to 384 * set the firmware into virtual mode. 385 * 386 * When EFI_PARAVIRT is in force then we could not map runtime 387 * service memory region because we do not have direct access to it. 388 * However, runtime services are available through proxy functions 389 * (e.g. in case of Xen dom0 EFI implementation they call special 390 * hypercall which executes relevant EFI functions) and that is why 391 * they are always enabled. 392 */ 393 394 if (!efi_enabled(EFI_PARAVIRT)) { 395 if (efi_enabled(EFI_64BIT)) 396 rv = efi_runtime_init64(); 397 else 398 rv = efi_runtime_init32(); 399 400 if (rv) 401 return rv; 402 } 403 404 set_bit(EFI_RUNTIME_SERVICES, &efi.flags); 405 406 return 0; 407} 408 409static int __init efi_memmap_init(void) 410{ 411 if (efi_enabled(EFI_PARAVIRT)) 412 return 0; 413 414 /* Map the EFI memory map */ 415 memmap.map = early_memremap((unsigned long)memmap.phys_map, 416 memmap.nr_map * memmap.desc_size); 417 if (memmap.map == NULL) { 418 pr_err("Could not map the memory map!\n"); 419 return -ENOMEM; 420 } 421 memmap.map_end = memmap.map + (memmap.nr_map * memmap.desc_size); 422 423 if (add_efi_memmap) 424 do_add_efi_memmap(); 425 426 set_bit(EFI_MEMMAP, &efi.flags); 427 428 return 0; 429} 430 431void __init efi_init(void) 432{ 433 efi_char16_t *c16; 434 char vendor[100] = "unknown"; 435 int i = 0; 436 void *tmp; 437 438#ifdef CONFIG_X86_32 439 if (boot_params.efi_info.efi_systab_hi || 440 boot_params.efi_info.efi_memmap_hi) { 441 pr_info("Table located above 4GB, disabling EFI.\n"); 442 return; 443 } 444 efi_phys.systab = (efi_system_table_t *)boot_params.efi_info.efi_systab; 445#else 446 efi_phys.systab = (efi_system_table_t *) 447 (boot_params.efi_info.efi_systab | 448 ((__u64)boot_params.efi_info.efi_systab_hi<<32)); 449#endif 450 451 if (efi_systab_init(efi_phys.systab)) 452 return; 453 454 efi.config_table = (unsigned long)efi.systab->tables; 455 efi.fw_vendor = (unsigned long)efi.systab->fw_vendor; 456 efi.runtime = (unsigned long)efi.systab->runtime; 457 458 /* 459 * Show what we know for posterity 460 */ 461 c16 = tmp = early_memremap(efi.systab->fw_vendor, 2); 462 if (c16) { 463 for (i = 0; i < sizeof(vendor) - 1 && *c16; ++i) 464 vendor[i] = *c16++; 465 vendor[i] = '\0'; 466 } else 467 pr_err("Could not map the firmware vendor!\n"); 468 early_memunmap(tmp, 2); 469 470 pr_info("EFI v%u.%.02u by %s\n", 471 efi.systab->hdr.revision >> 16, 472 efi.systab->hdr.revision & 0xffff, vendor); 473 474 if (efi_reuse_config(efi.systab->tables, efi.systab->nr_tables)) 475 return; 476 477 if (efi_config_init(arch_tables)) 478 return; 479 480 /* 481 * Note: We currently don't support runtime services on an EFI 482 * that doesn't match the kernel 32/64-bit mode. 483 */ 484 485 if (!efi_runtime_supported()) 486 pr_info("No EFI runtime due to 32/64-bit mismatch with kernel\n"); 487 else { 488 if (efi_runtime_disabled() || efi_runtime_init()) 489 return; 490 } 491 if (efi_memmap_init()) 492 return; 493 494 print_efi_memmap(); 495} 496 497void __init efi_late_init(void) 498{ 499 efi_bgrt_init(); 500} 501 502void __init efi_set_executable(efi_memory_desc_t *md, bool executable) 503{ 504 u64 addr, npages; 505 506 addr = md->virt_addr; 507 npages = md->num_pages; 508 509 memrange_efi_to_native(&addr, &npages); 510 511 if (executable) 512 set_memory_x(addr, npages); 513 else 514 set_memory_nx(addr, npages); 515} 516 517void __init runtime_code_page_mkexec(void) 518{ 519 efi_memory_desc_t *md; 520 void *p; 521 522 /* Make EFI runtime service code area executable */ 523 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) { 524 md = p; 525 526 if (md->type != EFI_RUNTIME_SERVICES_CODE) 527 continue; 528 529 efi_set_executable(md, true); 530 } 531} 532 533void __init efi_memory_uc(u64 addr, unsigned long size) 534{ 535 unsigned long page_shift = 1UL << EFI_PAGE_SHIFT; 536 u64 npages; 537 538 npages = round_up(size, page_shift) / page_shift; 539 memrange_efi_to_native(&addr, &npages); 540 set_memory_uc(addr, npages); 541} 542 543void __init old_map_region(efi_memory_desc_t *md) 544{ 545 u64 start_pfn, end_pfn, end; 546 unsigned long size; 547 void *va; 548 549 start_pfn = PFN_DOWN(md->phys_addr); 550 size = md->num_pages << PAGE_SHIFT; 551 end = md->phys_addr + size; 552 end_pfn = PFN_UP(end); 553 554 if (pfn_range_is_mapped(start_pfn, end_pfn)) { 555 va = __va(md->phys_addr); 556 557 if (!(md->attribute & EFI_MEMORY_WB)) 558 efi_memory_uc((u64)(unsigned long)va, size); 559 } else 560 va = efi_ioremap(md->phys_addr, size, 561 md->type, md->attribute); 562 563 md->virt_addr = (u64) (unsigned long) va; 564 if (!va) 565 pr_err("ioremap of 0x%llX failed!\n", 566 (unsigned long long)md->phys_addr); 567} 568 569/* Merge contiguous regions of the same type and attribute */ 570static void __init efi_merge_regions(void) 571{ 572 void *p; 573 efi_memory_desc_t *md, *prev_md = NULL; 574 575 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) { 576 u64 prev_size; 577 md = p; 578 579 if (!prev_md) { 580 prev_md = md; 581 continue; 582 } 583 584 if (prev_md->type != md->type || 585 prev_md->attribute != md->attribute) { 586 prev_md = md; 587 continue; 588 } 589 590 prev_size = prev_md->num_pages << EFI_PAGE_SHIFT; 591 592 if (md->phys_addr == (prev_md->phys_addr + prev_size)) { 593 prev_md->num_pages += md->num_pages; 594 md->type = EFI_RESERVED_TYPE; 595 md->attribute = 0; 596 continue; 597 } 598 prev_md = md; 599 } 600} 601 602static void __init get_systab_virt_addr(efi_memory_desc_t *md) 603{ 604 unsigned long size; 605 u64 end, systab; 606 607 size = md->num_pages << EFI_PAGE_SHIFT; 608 end = md->phys_addr + size; 609 systab = (u64)(unsigned long)efi_phys.systab; 610 if (md->phys_addr <= systab && systab < end) { 611 systab += md->virt_addr - md->phys_addr; 612 efi.systab = (efi_system_table_t *)(unsigned long)systab; 613 } 614} 615 616static void __init save_runtime_map(void) 617{ 618#ifdef CONFIG_KEXEC 619 efi_memory_desc_t *md; 620 void *tmp, *p, *q = NULL; 621 int count = 0; 622 623 if (efi_enabled(EFI_OLD_MEMMAP)) 624 return; 625 626 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) { 627 md = p; 628 629 if (!(md->attribute & EFI_MEMORY_RUNTIME) || 630 (md->type == EFI_BOOT_SERVICES_CODE) || 631 (md->type == EFI_BOOT_SERVICES_DATA)) 632 continue; 633 tmp = krealloc(q, (count + 1) * memmap.desc_size, GFP_KERNEL); 634 if (!tmp) 635 goto out; 636 q = tmp; 637 638 memcpy(q + count * memmap.desc_size, md, memmap.desc_size); 639 count++; 640 } 641 642 efi_runtime_map_setup(q, count, memmap.desc_size); 643 return; 644 645out: 646 kfree(q); 647 pr_err("Error saving runtime map, efi runtime on kexec non-functional!!\n"); 648#endif 649} 650 651static void *realloc_pages(void *old_memmap, int old_shift) 652{ 653 void *ret; 654 655 ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1); 656 if (!ret) 657 goto out; 658 659 /* 660 * A first-time allocation doesn't have anything to copy. 661 */ 662 if (!old_memmap) 663 return ret; 664 665 memcpy(ret, old_memmap, PAGE_SIZE << old_shift); 666 667out: 668 free_pages((unsigned long)old_memmap, old_shift); 669 return ret; 670} 671 672/* 673 * Map the efi memory ranges of the runtime services and update new_mmap with 674 * virtual addresses. 675 */ 676static void * __init efi_map_regions(int *count, int *pg_shift) 677{ 678 void *p, *new_memmap = NULL; 679 unsigned long left = 0; 680 efi_memory_desc_t *md; 681 682 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) { 683 md = p; 684 if (!(md->attribute & EFI_MEMORY_RUNTIME)) { 685#ifdef CONFIG_X86_64 686 if (md->type != EFI_BOOT_SERVICES_CODE && 687 md->type != EFI_BOOT_SERVICES_DATA) 688#endif 689 continue; 690 } 691 692 efi_map_region(md); 693 get_systab_virt_addr(md); 694 695 if (left < memmap.desc_size) { 696 new_memmap = realloc_pages(new_memmap, *pg_shift); 697 if (!new_memmap) 698 return NULL; 699 700 left += PAGE_SIZE << *pg_shift; 701 (*pg_shift)++; 702 } 703 704 memcpy(new_memmap + (*count * memmap.desc_size), md, 705 memmap.desc_size); 706 707 left -= memmap.desc_size; 708 (*count)++; 709 } 710 711 return new_memmap; 712} 713 714static void __init kexec_enter_virtual_mode(void) 715{ 716#ifdef CONFIG_KEXEC 717 efi_memory_desc_t *md; 718 void *p; 719 720 efi.systab = NULL; 721 722 /* 723 * We don't do virtual mode, since we don't do runtime services, on 724 * non-native EFI 725 */ 726 if (!efi_is_native()) { 727 efi_unmap_memmap(); 728 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags); 729 return; 730 } 731 732 /* 733 * Map efi regions which were passed via setup_data. The virt_addr is a 734 * fixed addr which was used in first kernel of a kexec boot. 735 */ 736 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) { 737 md = p; 738 efi_map_region_fixed(md); /* FIXME: add error handling */ 739 get_systab_virt_addr(md); 740 } 741 742 save_runtime_map(); 743 744 BUG_ON(!efi.systab); 745 746 efi_sync_low_kernel_mappings(); 747 748 /* 749 * Now that EFI is in virtual mode, update the function 750 * pointers in the runtime service table to the new virtual addresses. 751 * 752 * Call EFI services through wrapper functions. 753 */ 754 efi.runtime_version = efi_systab.hdr.revision; 755 756 efi_native_runtime_setup(); 757 758 efi.set_virtual_address_map = NULL; 759 760 if (efi_enabled(EFI_OLD_MEMMAP) && (__supported_pte_mask & _PAGE_NX)) 761 runtime_code_page_mkexec(); 762 763 /* clean DUMMY object */ 764 efi_delete_dummy_variable(); 765#endif 766} 767 768/* 769 * This function will switch the EFI runtime services to virtual mode. 770 * Essentially, we look through the EFI memmap and map every region that 771 * has the runtime attribute bit set in its memory descriptor into the 772 * ->trampoline_pgd page table using a top-down VA allocation scheme. 773 * 774 * The old method which used to update that memory descriptor with the 775 * virtual address obtained from ioremap() is still supported when the 776 * kernel is booted with efi=old_map on its command line. Same old 777 * method enabled the runtime services to be called without having to 778 * thunk back into physical mode for every invocation. 779 * 780 * The new method does a pagetable switch in a preemption-safe manner 781 * so that we're in a different address space when calling a runtime 782 * function. For function arguments passing we do copy the PGDs of the 783 * kernel page table into ->trampoline_pgd prior to each call. 784 * 785 * Specially for kexec boot, efi runtime maps in previous kernel should 786 * be passed in via setup_data. In that case runtime ranges will be mapped 787 * to the same virtual addresses as the first kernel, see 788 * kexec_enter_virtual_mode(). 789 */ 790static void __init __efi_enter_virtual_mode(void) 791{ 792 int count = 0, pg_shift = 0; 793 void *new_memmap = NULL; 794 efi_status_t status; 795 796 efi.systab = NULL; 797 798 efi_merge_regions(); 799 new_memmap = efi_map_regions(&count, &pg_shift); 800 if (!new_memmap) { 801 pr_err("Error reallocating memory, EFI runtime non-functional!\n"); 802 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags); 803 return; 804 } 805 806 save_runtime_map(); 807 808 BUG_ON(!efi.systab); 809 810 if (efi_setup_page_tables(__pa(new_memmap), 1 << pg_shift)) { 811 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags); 812 return; 813 } 814 815 efi_sync_low_kernel_mappings(); 816 efi_dump_pagetable(); 817 818 if (efi_is_native()) { 819 status = phys_efi_set_virtual_address_map( 820 memmap.desc_size * count, 821 memmap.desc_size, 822 memmap.desc_version, 823 (efi_memory_desc_t *)__pa(new_memmap)); 824 } else { 825 status = efi_thunk_set_virtual_address_map( 826 efi_phys.set_virtual_address_map, 827 memmap.desc_size * count, 828 memmap.desc_size, 829 memmap.desc_version, 830 (efi_memory_desc_t *)__pa(new_memmap)); 831 } 832 833 if (status != EFI_SUCCESS) { 834 pr_alert("Unable to switch EFI into virtual mode (status=%lx)!\n", 835 status); 836 panic("EFI call to SetVirtualAddressMap() failed!"); 837 } 838 839 /* 840 * Now that EFI is in virtual mode, update the function 841 * pointers in the runtime service table to the new virtual addresses. 842 * 843 * Call EFI services through wrapper functions. 844 */ 845 efi.runtime_version = efi_systab.hdr.revision; 846 847 if (efi_is_native()) 848 efi_native_runtime_setup(); 849 else 850 efi_thunk_runtime_setup(); 851 852 efi.set_virtual_address_map = NULL; 853 854 efi_runtime_mkexec(); 855 856 /* 857 * We mapped the descriptor array into the EFI pagetable above but we're 858 * not unmapping it here. Here's why: 859 * 860 * We're copying select PGDs from the kernel page table to the EFI page 861 * table and when we do so and make changes to those PGDs like unmapping 862 * stuff from them, those changes appear in the kernel page table and we 863 * go boom. 864 * 865 * From setup_real_mode(): 866 * 867 * ... 868 * trampoline_pgd[0] = init_level4_pgt[pgd_index(__PAGE_OFFSET)].pgd; 869 * 870 * In this particular case, our allocation is in PGD 0 of the EFI page 871 * table but we've copied that PGD from PGD[272] of the EFI page table: 872 * 873 * pgd_index(__PAGE_OFFSET = 0xffff880000000000) = 272 874 * 875 * where the direct memory mapping in kernel space is. 876 * 877 * new_memmap's VA comes from that direct mapping and thus clearing it, 878 * it would get cleared in the kernel page table too. 879 * 880 * efi_cleanup_page_tables(__pa(new_memmap), 1 << pg_shift); 881 */ 882 free_pages((unsigned long)new_memmap, pg_shift); 883 884 /* clean DUMMY object */ 885 efi_delete_dummy_variable(); 886} 887 888void __init efi_enter_virtual_mode(void) 889{ 890 if (efi_enabled(EFI_PARAVIRT)) 891 return; 892 893 if (efi_setup) 894 kexec_enter_virtual_mode(); 895 else 896 __efi_enter_virtual_mode(); 897} 898 899/* 900 * Convenience functions to obtain memory types and attributes 901 */ 902u32 efi_mem_type(unsigned long phys_addr) 903{ 904 efi_memory_desc_t *md; 905 void *p; 906 907 if (!efi_enabled(EFI_MEMMAP)) 908 return 0; 909 910 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) { 911 md = p; 912 if ((md->phys_addr <= phys_addr) && 913 (phys_addr < (md->phys_addr + 914 (md->num_pages << EFI_PAGE_SHIFT)))) 915 return md->type; 916 } 917 return 0; 918} 919 920u64 efi_mem_attributes(unsigned long phys_addr) 921{ 922 efi_memory_desc_t *md; 923 void *p; 924 925 if (!efi_enabled(EFI_MEMMAP)) 926 return 0; 927 928 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) { 929 md = p; 930 if ((md->phys_addr <= phys_addr) && 931 (phys_addr < (md->phys_addr + 932 (md->num_pages << EFI_PAGE_SHIFT)))) 933 return md->attribute; 934 } 935 return 0; 936} 937 938static int __init arch_parse_efi_cmdline(char *str) 939{ 940 if (parse_option_str(str, "old_map")) 941 set_bit(EFI_OLD_MEMMAP, &efi.flags); 942 943 return 0; 944} 945early_param("efi", arch_parse_efi_cmdline); 946