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