1 // SPDX-License-Identifier: GPL-2.0-only 2 #define pr_fmt(fmt) "efi: " fmt 3 4 #include <linux/init.h> 5 #include <linux/kernel.h> 6 #include <linux/string.h> 7 #include <linux/time.h> 8 #include <linux/types.h> 9 #include <linux/efi.h> 10 #include <linux/slab.h> 11 #include <linux/memblock.h> 12 #include <linux/acpi.h> 13 #include <linux/dmi.h> 14 15 #include <asm/e820/api.h> 16 #include <asm/efi.h> 17 #include <asm/uv/uv.h> 18 #include <asm/cpu_device_id.h> 19 #include <asm/realmode.h> 20 #include <asm/reboot.h> 21 22 #define EFI_MIN_RESERVE 5120 23 24 #define EFI_DUMMY_GUID \ 25 EFI_GUID(0x4424ac57, 0xbe4b, 0x47dd, 0x9e, 0x97, 0xed, 0x50, 0xf0, 0x9f, 0x92, 0xa9) 26 27 #define QUARK_CSH_SIGNATURE 0x5f435348 /* _CSH */ 28 #define QUARK_SECURITY_HEADER_SIZE 0x400 29 30 /* 31 * Header prepended to the standard EFI capsule on Quark systems the are based 32 * on Intel firmware BSP. 33 * @csh_signature: Unique identifier to sanity check signed module 34 * presence ("_CSH"). 35 * @version: Current version of CSH used. Should be one for Quark A0. 36 * @modulesize: Size of the entire module including the module header 37 * and payload. 38 * @security_version_number_index: Index of SVN to use for validation of signed 39 * module. 40 * @security_version_number: Used to prevent against roll back of modules. 41 * @rsvd_module_id: Currently unused for Clanton (Quark). 42 * @rsvd_module_vendor: Vendor Identifier. For Intel products value is 43 * 0x00008086. 44 * @rsvd_date: BCD representation of build date as yyyymmdd, where 45 * yyyy=4 digit year, mm=1-12, dd=1-31. 46 * @headersize: Total length of the header including including any 47 * padding optionally added by the signing tool. 48 * @hash_algo: What Hash is used in the module signing. 49 * @cryp_algo: What Crypto is used in the module signing. 50 * @keysize: Total length of the key data including including any 51 * padding optionally added by the signing tool. 52 * @signaturesize: Total length of the signature including including any 53 * padding optionally added by the signing tool. 54 * @rsvd_next_header: 32-bit pointer to the next Secure Boot Module in the 55 * chain, if there is a next header. 56 * @rsvd: Reserved, padding structure to required size. 57 * 58 * See also QuartSecurityHeader_t in 59 * Quark_EDKII_v1.2.1.1/QuarkPlatformPkg/Include/QuarkBootRom.h 60 * from https://downloadcenter.intel.com/download/23197/Intel-Quark-SoC-X1000-Board-Support-Package-BSP 61 */ 62 struct quark_security_header { 63 u32 csh_signature; 64 u32 version; 65 u32 modulesize; 66 u32 security_version_number_index; 67 u32 security_version_number; 68 u32 rsvd_module_id; 69 u32 rsvd_module_vendor; 70 u32 rsvd_date; 71 u32 headersize; 72 u32 hash_algo; 73 u32 cryp_algo; 74 u32 keysize; 75 u32 signaturesize; 76 u32 rsvd_next_header; 77 u32 rsvd[2]; 78 }; 79 80 static const efi_char16_t efi_dummy_name[] = L"DUMMY"; 81 82 static bool efi_no_storage_paranoia; 83 84 /* 85 * Some firmware implementations refuse to boot if there's insufficient 86 * space in the variable store. The implementation of garbage collection 87 * in some FW versions causes stale (deleted) variables to take up space 88 * longer than intended and space is only freed once the store becomes 89 * almost completely full. 90 * 91 * Enabling this option disables the space checks in 92 * efi_query_variable_store() and forces garbage collection. 93 * 94 * Only enable this option if deleting EFI variables does not free up 95 * space in your variable store, e.g. if despite deleting variables 96 * you're unable to create new ones. 97 */ 98 static int __init setup_storage_paranoia(char *arg) 99 { 100 efi_no_storage_paranoia = true; 101 return 0; 102 } 103 early_param("efi_no_storage_paranoia", setup_storage_paranoia); 104 105 /* 106 * Deleting the dummy variable which kicks off garbage collection 107 */ 108 void efi_delete_dummy_variable(void) 109 { 110 efi.set_variable_nonblocking((efi_char16_t *)efi_dummy_name, 111 &EFI_DUMMY_GUID, 112 EFI_VARIABLE_NON_VOLATILE | 113 EFI_VARIABLE_BOOTSERVICE_ACCESS | 114 EFI_VARIABLE_RUNTIME_ACCESS, 0, NULL); 115 } 116 117 /* 118 * In the nonblocking case we do not attempt to perform garbage 119 * collection if we do not have enough free space. Rather, we do the 120 * bare minimum check and give up immediately if the available space 121 * is below EFI_MIN_RESERVE. 122 * 123 * This function is intended to be small and simple because it is 124 * invoked from crash handler paths. 125 */ 126 static efi_status_t 127 query_variable_store_nonblocking(u32 attributes, unsigned long size) 128 { 129 efi_status_t status; 130 u64 storage_size, remaining_size, max_size; 131 132 status = efi.query_variable_info_nonblocking(attributes, &storage_size, 133 &remaining_size, 134 &max_size); 135 if (status != EFI_SUCCESS) 136 return status; 137 138 if (remaining_size - size < EFI_MIN_RESERVE) 139 return EFI_OUT_OF_RESOURCES; 140 141 return EFI_SUCCESS; 142 } 143 144 /* 145 * Some firmware implementations refuse to boot if there's insufficient space 146 * in the variable store. Ensure that we never use more than a safe limit. 147 * 148 * Return EFI_SUCCESS if it is safe to write 'size' bytes to the variable 149 * store. 150 */ 151 efi_status_t efi_query_variable_store(u32 attributes, unsigned long size, 152 bool nonblocking) 153 { 154 efi_status_t status; 155 u64 storage_size, remaining_size, max_size; 156 157 if (!(attributes & EFI_VARIABLE_NON_VOLATILE)) 158 return 0; 159 160 if (nonblocking) 161 return query_variable_store_nonblocking(attributes, size); 162 163 status = efi.query_variable_info(attributes, &storage_size, 164 &remaining_size, &max_size); 165 if (status != EFI_SUCCESS) 166 return status; 167 168 /* 169 * We account for that by refusing the write if permitting it would 170 * reduce the available space to under 5KB. This figure was provided by 171 * Samsung, so should be safe. 172 */ 173 if ((remaining_size - size < EFI_MIN_RESERVE) && 174 !efi_no_storage_paranoia) { 175 176 /* 177 * Triggering garbage collection may require that the firmware 178 * generate a real EFI_OUT_OF_RESOURCES error. We can force 179 * that by attempting to use more space than is available. 180 */ 181 unsigned long dummy_size = remaining_size + 1024; 182 void *dummy = kzalloc(dummy_size, GFP_KERNEL); 183 184 if (!dummy) 185 return EFI_OUT_OF_RESOURCES; 186 187 status = efi.set_variable((efi_char16_t *)efi_dummy_name, 188 &EFI_DUMMY_GUID, 189 EFI_VARIABLE_NON_VOLATILE | 190 EFI_VARIABLE_BOOTSERVICE_ACCESS | 191 EFI_VARIABLE_RUNTIME_ACCESS, 192 dummy_size, dummy); 193 194 if (status == EFI_SUCCESS) { 195 /* 196 * This should have failed, so if it didn't make sure 197 * that we delete it... 198 */ 199 efi_delete_dummy_variable(); 200 } 201 202 kfree(dummy); 203 204 /* 205 * The runtime code may now have triggered a garbage collection 206 * run, so check the variable info again 207 */ 208 status = efi.query_variable_info(attributes, &storage_size, 209 &remaining_size, &max_size); 210 211 if (status != EFI_SUCCESS) 212 return status; 213 214 /* 215 * There still isn't enough room, so return an error 216 */ 217 if (remaining_size - size < EFI_MIN_RESERVE) 218 return EFI_OUT_OF_RESOURCES; 219 } 220 221 return EFI_SUCCESS; 222 } 223 EXPORT_SYMBOL_GPL(efi_query_variable_store); 224 225 /* 226 * The UEFI specification makes it clear that the operating system is 227 * free to do whatever it wants with boot services code after 228 * ExitBootServices() has been called. Ignoring this recommendation a 229 * significant bunch of EFI implementations continue calling into boot 230 * services code (SetVirtualAddressMap). In order to work around such 231 * buggy implementations we reserve boot services region during EFI 232 * init and make sure it stays executable. Then, after 233 * SetVirtualAddressMap(), it is discarded. 234 * 235 * However, some boot services regions contain data that is required 236 * by drivers, so we need to track which memory ranges can never be 237 * freed. This is done by tagging those regions with the 238 * EFI_MEMORY_RUNTIME attribute. 239 * 240 * Any driver that wants to mark a region as reserved must use 241 * efi_mem_reserve() which will insert a new EFI memory descriptor 242 * into efi.memmap (splitting existing regions if necessary) and tag 243 * it with EFI_MEMORY_RUNTIME. 244 */ 245 void __init efi_arch_mem_reserve(phys_addr_t addr, u64 size) 246 { 247 struct efi_memory_map_data data = { 0 }; 248 struct efi_mem_range mr; 249 efi_memory_desc_t md; 250 int num_entries; 251 void *new; 252 253 if (efi_mem_desc_lookup(addr, &md) || 254 md.type != EFI_BOOT_SERVICES_DATA) { 255 pr_err("Failed to lookup EFI memory descriptor for %pa\n", &addr); 256 return; 257 } 258 259 if (addr + size > md.phys_addr + (md.num_pages << EFI_PAGE_SHIFT)) { 260 pr_err("Region spans EFI memory descriptors, %pa\n", &addr); 261 return; 262 } 263 264 size += addr % EFI_PAGE_SIZE; 265 size = round_up(size, EFI_PAGE_SIZE); 266 addr = round_down(addr, EFI_PAGE_SIZE); 267 268 mr.range.start = addr; 269 mr.range.end = addr + size - 1; 270 mr.attribute = md.attribute | EFI_MEMORY_RUNTIME; 271 272 num_entries = efi_memmap_split_count(&md, &mr.range); 273 num_entries += efi.memmap.nr_map; 274 275 if (efi_memmap_alloc(num_entries, &data) != 0) { 276 pr_err("Could not allocate boot services memmap\n"); 277 return; 278 } 279 280 new = early_memremap(data.phys_map, data.size); 281 if (!new) { 282 pr_err("Failed to map new boot services memmap\n"); 283 return; 284 } 285 286 efi_memmap_insert(&efi.memmap, new, &mr); 287 early_memunmap(new, data.size); 288 289 efi_memmap_install(&data); 290 e820__range_update(addr, size, E820_TYPE_RAM, E820_TYPE_RESERVED); 291 e820__update_table(e820_table); 292 } 293 294 /* 295 * Helper function for efi_reserve_boot_services() to figure out if we 296 * can free regions in efi_free_boot_services(). 297 * 298 * Use this function to ensure we do not free regions owned by somebody 299 * else. We must only reserve (and then free) regions: 300 * 301 * - Not within any part of the kernel 302 * - Not the BIOS reserved area (E820_TYPE_RESERVED, E820_TYPE_NVS, etc) 303 */ 304 static __init bool can_free_region(u64 start, u64 size) 305 { 306 if (start + size > __pa_symbol(_text) && start <= __pa_symbol(_end)) 307 return false; 308 309 if (!e820__mapped_all(start, start+size, E820_TYPE_RAM)) 310 return false; 311 312 return true; 313 } 314 315 void __init efi_reserve_boot_services(void) 316 { 317 efi_memory_desc_t *md; 318 319 if (!efi_enabled(EFI_MEMMAP)) 320 return; 321 322 for_each_efi_memory_desc(md) { 323 u64 start = md->phys_addr; 324 u64 size = md->num_pages << EFI_PAGE_SHIFT; 325 bool already_reserved; 326 327 if (md->type != EFI_BOOT_SERVICES_CODE && 328 md->type != EFI_BOOT_SERVICES_DATA) 329 continue; 330 331 already_reserved = memblock_is_region_reserved(start, size); 332 333 /* 334 * Because the following memblock_reserve() is paired 335 * with memblock_free_late() for this region in 336 * efi_free_boot_services(), we must be extremely 337 * careful not to reserve, and subsequently free, 338 * critical regions of memory (like the kernel image) or 339 * those regions that somebody else has already 340 * reserved. 341 * 342 * A good example of a critical region that must not be 343 * freed is page zero (first 4Kb of memory), which may 344 * contain boot services code/data but is marked 345 * E820_TYPE_RESERVED by trim_bios_range(). 346 */ 347 if (!already_reserved) { 348 memblock_reserve(start, size); 349 350 /* 351 * If we are the first to reserve the region, no 352 * one else cares about it. We own it and can 353 * free it later. 354 */ 355 if (can_free_region(start, size)) 356 continue; 357 } 358 359 /* 360 * We don't own the region. We must not free it. 361 * 362 * Setting this bit for a boot services region really 363 * doesn't make sense as far as the firmware is 364 * concerned, but it does provide us with a way to tag 365 * those regions that must not be paired with 366 * memblock_free_late(). 367 */ 368 md->attribute |= EFI_MEMORY_RUNTIME; 369 } 370 } 371 372 /* 373 * Apart from having VA mappings for EFI boot services code/data regions, 374 * (duplicate) 1:1 mappings were also created as a quirk for buggy firmware. So, 375 * unmap both 1:1 and VA mappings. 376 */ 377 static void __init efi_unmap_pages(efi_memory_desc_t *md) 378 { 379 pgd_t *pgd = efi_mm.pgd; 380 u64 pa = md->phys_addr; 381 u64 va = md->virt_addr; 382 383 /* 384 * To Do: Remove this check after adding functionality to unmap EFI boot 385 * services code/data regions from direct mapping area because the UV1 386 * memory map maps EFI regions in swapper_pg_dir. 387 */ 388 if (efi_have_uv1_memmap()) 389 return; 390 391 /* 392 * EFI mixed mode has all RAM mapped to access arguments while making 393 * EFI runtime calls, hence don't unmap EFI boot services code/data 394 * regions. 395 */ 396 if (efi_is_mixed()) 397 return; 398 399 if (kernel_unmap_pages_in_pgd(pgd, pa, md->num_pages)) 400 pr_err("Failed to unmap 1:1 mapping for 0x%llx\n", pa); 401 402 if (kernel_unmap_pages_in_pgd(pgd, va, md->num_pages)) 403 pr_err("Failed to unmap VA mapping for 0x%llx\n", va); 404 } 405 406 void __init efi_free_boot_services(void) 407 { 408 struct efi_memory_map_data data = { 0 }; 409 efi_memory_desc_t *md; 410 int num_entries = 0; 411 void *new, *new_md; 412 413 for_each_efi_memory_desc(md) { 414 unsigned long long start = md->phys_addr; 415 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT; 416 size_t rm_size; 417 418 if (md->type != EFI_BOOT_SERVICES_CODE && 419 md->type != EFI_BOOT_SERVICES_DATA) { 420 num_entries++; 421 continue; 422 } 423 424 /* Do not free, someone else owns it: */ 425 if (md->attribute & EFI_MEMORY_RUNTIME) { 426 num_entries++; 427 continue; 428 } 429 430 /* 431 * Before calling set_virtual_address_map(), EFI boot services 432 * code/data regions were mapped as a quirk for buggy firmware. 433 * Unmap them from efi_pgd before freeing them up. 434 */ 435 efi_unmap_pages(md); 436 437 /* 438 * Nasty quirk: if all sub-1MB memory is used for boot 439 * services, we can get here without having allocated the 440 * real mode trampoline. It's too late to hand boot services 441 * memory back to the memblock allocator, so instead 442 * try to manually allocate the trampoline if needed. 443 * 444 * I've seen this on a Dell XPS 13 9350 with firmware 445 * 1.4.4 with SGX enabled booting Linux via Fedora 24's 446 * grub2-efi on a hard disk. (And no, I don't know why 447 * this happened, but Linux should still try to boot rather 448 * panicing early.) 449 */ 450 rm_size = real_mode_size_needed(); 451 if (rm_size && (start + rm_size) < (1<<20) && size >= rm_size) { 452 set_real_mode_mem(start); 453 start += rm_size; 454 size -= rm_size; 455 } 456 457 memblock_free_late(start, size); 458 } 459 460 if (!num_entries) 461 return; 462 463 if (efi_memmap_alloc(num_entries, &data) != 0) { 464 pr_err("Failed to allocate new EFI memmap\n"); 465 return; 466 } 467 468 new = memremap(data.phys_map, data.size, MEMREMAP_WB); 469 if (!new) { 470 pr_err("Failed to map new EFI memmap\n"); 471 return; 472 } 473 474 /* 475 * Build a new EFI memmap that excludes any boot services 476 * regions that are not tagged EFI_MEMORY_RUNTIME, since those 477 * regions have now been freed. 478 */ 479 new_md = new; 480 for_each_efi_memory_desc(md) { 481 if (!(md->attribute & EFI_MEMORY_RUNTIME) && 482 (md->type == EFI_BOOT_SERVICES_CODE || 483 md->type == EFI_BOOT_SERVICES_DATA)) 484 continue; 485 486 memcpy(new_md, md, efi.memmap.desc_size); 487 new_md += efi.memmap.desc_size; 488 } 489 490 memunmap(new); 491 492 if (efi_memmap_install(&data) != 0) { 493 pr_err("Could not install new EFI memmap\n"); 494 return; 495 } 496 } 497 498 /* 499 * A number of config table entries get remapped to virtual addresses 500 * after entering EFI virtual mode. However, the kexec kernel requires 501 * their physical addresses therefore we pass them via setup_data and 502 * correct those entries to their respective physical addresses here. 503 * 504 * Currently only handles smbios which is necessary for some firmware 505 * implementation. 506 */ 507 int __init efi_reuse_config(u64 tables, int nr_tables) 508 { 509 int i, sz, ret = 0; 510 void *p, *tablep; 511 struct efi_setup_data *data; 512 513 if (nr_tables == 0) 514 return 0; 515 516 if (!efi_setup) 517 return 0; 518 519 if (!efi_enabled(EFI_64BIT)) 520 return 0; 521 522 data = early_memremap(efi_setup, sizeof(*data)); 523 if (!data) { 524 ret = -ENOMEM; 525 goto out; 526 } 527 528 if (!data->smbios) 529 goto out_memremap; 530 531 sz = sizeof(efi_config_table_64_t); 532 533 p = tablep = early_memremap(tables, nr_tables * sz); 534 if (!p) { 535 pr_err("Could not map Configuration table!\n"); 536 ret = -ENOMEM; 537 goto out_memremap; 538 } 539 540 for (i = 0; i < efi.systab->nr_tables; i++) { 541 efi_guid_t guid; 542 543 guid = ((efi_config_table_64_t *)p)->guid; 544 545 if (!efi_guidcmp(guid, SMBIOS_TABLE_GUID)) 546 ((efi_config_table_64_t *)p)->table = data->smbios; 547 p += sz; 548 } 549 early_memunmap(tablep, nr_tables * sz); 550 551 out_memremap: 552 early_memunmap(data, sizeof(*data)); 553 out: 554 return ret; 555 } 556 557 static const struct dmi_system_id sgi_uv1_dmi[] __initconst = { 558 { NULL, "SGI UV1", 559 { DMI_MATCH(DMI_PRODUCT_NAME, "Stoutland Platform"), 560 DMI_MATCH(DMI_PRODUCT_VERSION, "1.0"), 561 DMI_MATCH(DMI_BIOS_VENDOR, "SGI.COM"), 562 } 563 }, 564 { } /* NULL entry stops DMI scanning */ 565 }; 566 567 void __init efi_apply_memmap_quirks(void) 568 { 569 /* 570 * Once setup is done earlier, unmap the EFI memory map on mismatched 571 * firmware/kernel architectures since there is no support for runtime 572 * services. 573 */ 574 if (!efi_runtime_supported()) { 575 pr_info("Setup done, disabling due to 32/64-bit mismatch\n"); 576 efi_memmap_unmap(); 577 } 578 579 /* UV2+ BIOS has a fix for this issue. UV1 still needs the quirk. */ 580 if (dmi_check_system(sgi_uv1_dmi)) { 581 if (IS_ENABLED(CONFIG_X86_UV)) { 582 set_bit(EFI_UV1_MEMMAP, &efi.flags); 583 } else { 584 pr_warn("EFI runtime disabled, needs CONFIG_X86_UV=y on UV1\n"); 585 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags); 586 efi_memmap_unmap(); 587 } 588 } 589 } 590 591 /* 592 * For most modern platforms the preferred method of powering off is via 593 * ACPI. However, there are some that are known to require the use of 594 * EFI runtime services and for which ACPI does not work at all. 595 * 596 * Using EFI is a last resort, to be used only if no other option 597 * exists. 598 */ 599 bool efi_reboot_required(void) 600 { 601 if (!acpi_gbl_reduced_hardware) 602 return false; 603 604 efi_reboot_quirk_mode = EFI_RESET_WARM; 605 return true; 606 } 607 608 bool efi_poweroff_required(void) 609 { 610 return acpi_gbl_reduced_hardware || acpi_no_s5; 611 } 612 613 #ifdef CONFIG_EFI_CAPSULE_QUIRK_QUARK_CSH 614 615 static int qrk_capsule_setup_info(struct capsule_info *cap_info, void **pkbuff, 616 size_t hdr_bytes) 617 { 618 struct quark_security_header *csh = *pkbuff; 619 620 /* Only process data block that is larger than the security header */ 621 if (hdr_bytes < sizeof(struct quark_security_header)) 622 return 0; 623 624 if (csh->csh_signature != QUARK_CSH_SIGNATURE || 625 csh->headersize != QUARK_SECURITY_HEADER_SIZE) 626 return 1; 627 628 /* Only process data block if EFI header is included */ 629 if (hdr_bytes < QUARK_SECURITY_HEADER_SIZE + 630 sizeof(efi_capsule_header_t)) 631 return 0; 632 633 pr_debug("Quark security header detected\n"); 634 635 if (csh->rsvd_next_header != 0) { 636 pr_err("multiple Quark security headers not supported\n"); 637 return -EINVAL; 638 } 639 640 *pkbuff += csh->headersize; 641 cap_info->total_size = csh->headersize; 642 643 /* 644 * Update the first page pointer to skip over the CSH header. 645 */ 646 cap_info->phys[0] += csh->headersize; 647 648 /* 649 * cap_info->capsule should point at a virtual mapping of the entire 650 * capsule, starting at the capsule header. Our image has the Quark 651 * security header prepended, so we cannot rely on the default vmap() 652 * mapping created by the generic capsule code. 653 * Given that the Quark firmware does not appear to care about the 654 * virtual mapping, let's just point cap_info->capsule at our copy 655 * of the capsule header. 656 */ 657 cap_info->capsule = &cap_info->header; 658 659 return 1; 660 } 661 662 #define ICPU(family, model, quirk_handler) \ 663 { X86_VENDOR_INTEL, family, model, X86_FEATURE_ANY, \ 664 (unsigned long)&quirk_handler } 665 666 static const struct x86_cpu_id efi_capsule_quirk_ids[] = { 667 ICPU(5, 9, qrk_capsule_setup_info), /* Intel Quark X1000 */ 668 { } 669 }; 670 671 int efi_capsule_setup_info(struct capsule_info *cap_info, void *kbuff, 672 size_t hdr_bytes) 673 { 674 int (*quirk_handler)(struct capsule_info *, void **, size_t); 675 const struct x86_cpu_id *id; 676 int ret; 677 678 if (hdr_bytes < sizeof(efi_capsule_header_t)) 679 return 0; 680 681 cap_info->total_size = 0; 682 683 id = x86_match_cpu(efi_capsule_quirk_ids); 684 if (id) { 685 /* 686 * The quirk handler is supposed to return 687 * - a value > 0 if the setup should continue, after advancing 688 * kbuff as needed 689 * - 0 if not enough hdr_bytes are available yet 690 * - a negative error code otherwise 691 */ 692 quirk_handler = (typeof(quirk_handler))id->driver_data; 693 ret = quirk_handler(cap_info, &kbuff, hdr_bytes); 694 if (ret <= 0) 695 return ret; 696 } 697 698 memcpy(&cap_info->header, kbuff, sizeof(cap_info->header)); 699 700 cap_info->total_size += cap_info->header.imagesize; 701 702 return __efi_capsule_setup_info(cap_info); 703 } 704 705 #endif 706 707 /* 708 * If any access by any efi runtime service causes a page fault, then, 709 * 1. If it's efi_reset_system(), reboot through BIOS. 710 * 2. If any other efi runtime service, then 711 * a. Return error status to the efi caller process. 712 * b. Disable EFI Runtime Services forever and 713 * c. Freeze efi_rts_wq and schedule new process. 714 * 715 * @return: Returns, if the page fault is not handled. This function 716 * will never return if the page fault is handled successfully. 717 */ 718 void efi_recover_from_page_fault(unsigned long phys_addr) 719 { 720 if (!IS_ENABLED(CONFIG_X86_64)) 721 return; 722 723 /* 724 * Make sure that an efi runtime service caused the page fault. 725 * "efi_mm" cannot be used to check if the page fault had occurred 726 * in the firmware context because the UV1 memmap doesn't use efi_pgd. 727 */ 728 if (efi_rts_work.efi_rts_id == EFI_NONE) 729 return; 730 731 /* 732 * Address range 0x0000 - 0x0fff is always mapped in the efi_pgd, so 733 * page faulting on these addresses isn't expected. 734 */ 735 if (phys_addr <= 0x0fff) 736 return; 737 738 /* 739 * Print stack trace as it might be useful to know which EFI Runtime 740 * Service is buggy. 741 */ 742 WARN(1, FW_BUG "Page fault caused by firmware at PA: 0x%lx\n", 743 phys_addr); 744 745 /* 746 * Buggy efi_reset_system() is handled differently from other EFI 747 * Runtime Services as it doesn't use efi_rts_wq. Although, 748 * native_machine_emergency_restart() says that machine_real_restart() 749 * could fail, it's better not to compilcate this fault handler 750 * because this case occurs *very* rarely and hence could be improved 751 * on a need by basis. 752 */ 753 if (efi_rts_work.efi_rts_id == EFI_RESET_SYSTEM) { 754 pr_info("efi_reset_system() buggy! Reboot through BIOS\n"); 755 machine_real_restart(MRR_BIOS); 756 return; 757 } 758 759 /* 760 * Before calling EFI Runtime Service, the kernel has switched the 761 * calling process to efi_mm. Hence, switch back to task_mm. 762 */ 763 arch_efi_call_virt_teardown(); 764 765 /* Signal error status to the efi caller process */ 766 efi_rts_work.status = EFI_ABORTED; 767 complete(&efi_rts_work.efi_rts_comp); 768 769 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags); 770 pr_info("Froze efi_rts_wq and disabled EFI Runtime Services\n"); 771 772 /* 773 * Call schedule() in an infinite loop, so that any spurious wake ups 774 * will never run efi_rts_wq again. 775 */ 776 for (;;) { 777 set_current_state(TASK_IDLE); 778 schedule(); 779 } 780 781 return; 782 } 783