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 */
setup_storage_paranoia(char * arg)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 */
efi_delete_dummy_variable(void)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
efivar_reserved_space(void)117 u64 efivar_reserved_space(void)
118 {
119 if (efi_no_storage_paranoia)
120 return 0;
121 return EFI_MIN_RESERVE;
122 }
123 EXPORT_SYMBOL_GPL(efivar_reserved_space);
124
125 /*
126 * In the nonblocking case we do not attempt to perform garbage
127 * collection if we do not have enough free space. Rather, we do the
128 * bare minimum check and give up immediately if the available space
129 * is below EFI_MIN_RESERVE.
130 *
131 * This function is intended to be small and simple because it is
132 * invoked from crash handler paths.
133 */
134 static efi_status_t
query_variable_store_nonblocking(u32 attributes,unsigned long size)135 query_variable_store_nonblocking(u32 attributes, unsigned long size)
136 {
137 efi_status_t status;
138 u64 storage_size, remaining_size, max_size;
139
140 status = efi.query_variable_info_nonblocking(attributes, &storage_size,
141 &remaining_size,
142 &max_size);
143 if (status != EFI_SUCCESS)
144 return status;
145
146 if (remaining_size - size < EFI_MIN_RESERVE)
147 return EFI_OUT_OF_RESOURCES;
148
149 return EFI_SUCCESS;
150 }
151
152 /*
153 * Some firmware implementations refuse to boot if there's insufficient space
154 * in the variable store. Ensure that we never use more than a safe limit.
155 *
156 * Return EFI_SUCCESS if it is safe to write 'size' bytes to the variable
157 * store.
158 */
efi_query_variable_store(u32 attributes,unsigned long size,bool nonblocking)159 efi_status_t efi_query_variable_store(u32 attributes, unsigned long size,
160 bool nonblocking)
161 {
162 efi_status_t status;
163 u64 storage_size, remaining_size, max_size;
164
165 if (!(attributes & EFI_VARIABLE_NON_VOLATILE))
166 return 0;
167
168 if (nonblocking)
169 return query_variable_store_nonblocking(attributes, size);
170
171 status = efi.query_variable_info(attributes, &storage_size,
172 &remaining_size, &max_size);
173 if (status != EFI_SUCCESS)
174 return status;
175
176 /*
177 * We account for that by refusing the write if permitting it would
178 * reduce the available space to under 5KB. This figure was provided by
179 * Samsung, so should be safe.
180 */
181 if ((remaining_size - size < EFI_MIN_RESERVE) &&
182 !efi_no_storage_paranoia) {
183
184 /*
185 * Triggering garbage collection may require that the firmware
186 * generate a real EFI_OUT_OF_RESOURCES error. We can force
187 * that by attempting to use more space than is available.
188 */
189 unsigned long dummy_size = remaining_size + 1024;
190 void *dummy = kzalloc(dummy_size, GFP_KERNEL);
191
192 if (!dummy)
193 return EFI_OUT_OF_RESOURCES;
194
195 status = efi.set_variable((efi_char16_t *)efi_dummy_name,
196 &EFI_DUMMY_GUID,
197 EFI_VARIABLE_NON_VOLATILE |
198 EFI_VARIABLE_BOOTSERVICE_ACCESS |
199 EFI_VARIABLE_RUNTIME_ACCESS,
200 dummy_size, dummy);
201
202 if (status == EFI_SUCCESS) {
203 /*
204 * This should have failed, so if it didn't make sure
205 * that we delete it...
206 */
207 efi_delete_dummy_variable();
208 }
209
210 kfree(dummy);
211
212 /*
213 * The runtime code may now have triggered a garbage collection
214 * run, so check the variable info again
215 */
216 status = efi.query_variable_info(attributes, &storage_size,
217 &remaining_size, &max_size);
218
219 if (status != EFI_SUCCESS)
220 return status;
221
222 /*
223 * There still isn't enough room, so return an error
224 */
225 if (remaining_size - size < EFI_MIN_RESERVE)
226 return EFI_OUT_OF_RESOURCES;
227 }
228
229 return EFI_SUCCESS;
230 }
231 EXPORT_SYMBOL_GPL(efi_query_variable_store);
232
233 /*
234 * The UEFI specification makes it clear that the operating system is
235 * free to do whatever it wants with boot services code after
236 * ExitBootServices() has been called. Ignoring this recommendation a
237 * significant bunch of EFI implementations continue calling into boot
238 * services code (SetVirtualAddressMap). In order to work around such
239 * buggy implementations we reserve boot services region during EFI
240 * init and make sure it stays executable. Then, after
241 * SetVirtualAddressMap(), it is discarded.
242 *
243 * However, some boot services regions contain data that is required
244 * by drivers, so we need to track which memory ranges can never be
245 * freed. This is done by tagging those regions with the
246 * EFI_MEMORY_RUNTIME attribute.
247 *
248 * Any driver that wants to mark a region as reserved must use
249 * efi_mem_reserve() which will insert a new EFI memory descriptor
250 * into efi.memmap (splitting existing regions if necessary) and tag
251 * it with EFI_MEMORY_RUNTIME.
252 */
efi_arch_mem_reserve(phys_addr_t addr,u64 size)253 void __init efi_arch_mem_reserve(phys_addr_t addr, u64 size)
254 {
255 struct efi_memory_map_data data = { 0 };
256 struct efi_mem_range mr;
257 efi_memory_desc_t md;
258 int num_entries;
259 void *new;
260
261 if (efi_mem_desc_lookup(addr, &md) ||
262 md.type != EFI_BOOT_SERVICES_DATA) {
263 pr_err("Failed to lookup EFI memory descriptor for %pa\n", &addr);
264 return;
265 }
266
267 if (addr + size > md.phys_addr + (md.num_pages << EFI_PAGE_SHIFT)) {
268 pr_err("Region spans EFI memory descriptors, %pa\n", &addr);
269 return;
270 }
271
272 size += addr % EFI_PAGE_SIZE;
273 size = round_up(size, EFI_PAGE_SIZE);
274 addr = round_down(addr, EFI_PAGE_SIZE);
275
276 mr.range.start = addr;
277 mr.range.end = addr + size - 1;
278 mr.attribute = md.attribute | EFI_MEMORY_RUNTIME;
279
280 num_entries = efi_memmap_split_count(&md, &mr.range);
281 num_entries += efi.memmap.nr_map;
282
283 if (efi_memmap_alloc(num_entries, &data) != 0) {
284 pr_err("Could not allocate boot services memmap\n");
285 return;
286 }
287
288 new = early_memremap_prot(data.phys_map, data.size,
289 pgprot_val(pgprot_encrypted(FIXMAP_PAGE_NORMAL)));
290 if (!new) {
291 pr_err("Failed to map new boot services memmap\n");
292 return;
293 }
294
295 efi_memmap_insert(&efi.memmap, new, &mr);
296 early_memunmap(new, data.size);
297
298 efi_memmap_install(&data);
299 e820__range_update(addr, size, E820_TYPE_RAM, E820_TYPE_RESERVED);
300 e820__update_table(e820_table);
301 }
302
303 /*
304 * Helper function for efi_reserve_boot_services() to figure out if we
305 * can free regions in efi_free_boot_services().
306 *
307 * Use this function to ensure we do not free regions owned by somebody
308 * else. We must only reserve (and then free) regions:
309 *
310 * - Not within any part of the kernel
311 * - Not the BIOS reserved area (E820_TYPE_RESERVED, E820_TYPE_NVS, etc)
312 */
can_free_region(u64 start,u64 size)313 static __init bool can_free_region(u64 start, u64 size)
314 {
315 if (start + size > __pa_symbol(_text) && start <= __pa_symbol(_end))
316 return false;
317
318 if (!e820__mapped_all(start, start+size, E820_TYPE_RAM))
319 return false;
320
321 return true;
322 }
323
efi_reserve_boot_services(void)324 void __init efi_reserve_boot_services(void)
325 {
326 efi_memory_desc_t *md;
327
328 if (!efi_enabled(EFI_MEMMAP))
329 return;
330
331 for_each_efi_memory_desc(md) {
332 u64 start = md->phys_addr;
333 u64 size = md->num_pages << EFI_PAGE_SHIFT;
334 bool already_reserved;
335
336 if (md->type != EFI_BOOT_SERVICES_CODE &&
337 md->type != EFI_BOOT_SERVICES_DATA)
338 continue;
339
340 already_reserved = memblock_is_region_reserved(start, size);
341
342 /*
343 * Because the following memblock_reserve() is paired
344 * with memblock_free_late() for this region in
345 * efi_free_boot_services(), we must be extremely
346 * careful not to reserve, and subsequently free,
347 * critical regions of memory (like the kernel image) or
348 * those regions that somebody else has already
349 * reserved.
350 *
351 * A good example of a critical region that must not be
352 * freed is page zero (first 4Kb of memory), which may
353 * contain boot services code/data but is marked
354 * E820_TYPE_RESERVED by trim_bios_range().
355 */
356 if (!already_reserved) {
357 memblock_reserve(start, size);
358
359 /*
360 * If we are the first to reserve the region, no
361 * one else cares about it. We own it and can
362 * free it later.
363 */
364 if (can_free_region(start, size))
365 continue;
366 }
367
368 /*
369 * We don't own the region. We must not free it.
370 *
371 * Setting this bit for a boot services region really
372 * doesn't make sense as far as the firmware is
373 * concerned, but it does provide us with a way to tag
374 * those regions that must not be paired with
375 * memblock_free_late().
376 */
377 md->attribute |= EFI_MEMORY_RUNTIME;
378 }
379 }
380
381 /*
382 * Apart from having VA mappings for EFI boot services code/data regions,
383 * (duplicate) 1:1 mappings were also created as a quirk for buggy firmware. So,
384 * unmap both 1:1 and VA mappings.
385 */
efi_unmap_pages(efi_memory_desc_t * md)386 static void __init efi_unmap_pages(efi_memory_desc_t *md)
387 {
388 pgd_t *pgd = efi_mm.pgd;
389 u64 pa = md->phys_addr;
390 u64 va = md->virt_addr;
391
392 /*
393 * EFI mixed mode has all RAM mapped to access arguments while making
394 * EFI runtime calls, hence don't unmap EFI boot services code/data
395 * regions.
396 */
397 if (efi_is_mixed())
398 return;
399
400 if (kernel_unmap_pages_in_pgd(pgd, pa, md->num_pages))
401 pr_err("Failed to unmap 1:1 mapping for 0x%llx\n", pa);
402
403 if (kernel_unmap_pages_in_pgd(pgd, va, md->num_pages))
404 pr_err("Failed to unmap VA mapping for 0x%llx\n", va);
405 }
406
efi_free_boot_services(void)407 void __init efi_free_boot_services(void)
408 {
409 struct efi_memory_map_data data = { 0 };
410 efi_memory_desc_t *md;
411 int num_entries = 0;
412 void *new, *new_md;
413
414 /* Keep all regions for /sys/kernel/debug/efi */
415 if (efi_enabled(EFI_DBG))
416 return;
417
418 for_each_efi_memory_desc(md) {
419 unsigned long long start = md->phys_addr;
420 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
421 size_t rm_size;
422
423 if (md->type != EFI_BOOT_SERVICES_CODE &&
424 md->type != EFI_BOOT_SERVICES_DATA) {
425 num_entries++;
426 continue;
427 }
428
429 /* Do not free, someone else owns it: */
430 if (md->attribute & EFI_MEMORY_RUNTIME) {
431 num_entries++;
432 continue;
433 }
434
435 /*
436 * Before calling set_virtual_address_map(), EFI boot services
437 * code/data regions were mapped as a quirk for buggy firmware.
438 * Unmap them from efi_pgd before freeing them up.
439 */
440 efi_unmap_pages(md);
441
442 /*
443 * Nasty quirk: if all sub-1MB memory is used for boot
444 * services, we can get here without having allocated the
445 * real mode trampoline. It's too late to hand boot services
446 * memory back to the memblock allocator, so instead
447 * try to manually allocate the trampoline if needed.
448 *
449 * I've seen this on a Dell XPS 13 9350 with firmware
450 * 1.4.4 with SGX enabled booting Linux via Fedora 24's
451 * grub2-efi on a hard disk. (And no, I don't know why
452 * this happened, but Linux should still try to boot rather
453 * panicking early.)
454 */
455 rm_size = real_mode_size_needed();
456 if (rm_size && (start + rm_size) < (1<<20) && size >= rm_size) {
457 set_real_mode_mem(start);
458 start += rm_size;
459 size -= rm_size;
460 }
461
462 /*
463 * Don't free memory under 1M for two reasons:
464 * - BIOS might clobber it
465 * - Crash kernel needs it to be reserved
466 */
467 if (start + size < SZ_1M)
468 continue;
469 if (start < SZ_1M) {
470 size -= (SZ_1M - start);
471 start = SZ_1M;
472 }
473
474 memblock_free_late(start, size);
475 }
476
477 if (!num_entries)
478 return;
479
480 if (efi_memmap_alloc(num_entries, &data) != 0) {
481 pr_err("Failed to allocate new EFI memmap\n");
482 return;
483 }
484
485 new = memremap(data.phys_map, data.size, MEMREMAP_WB);
486 if (!new) {
487 pr_err("Failed to map new EFI memmap\n");
488 return;
489 }
490
491 /*
492 * Build a new EFI memmap that excludes any boot services
493 * regions that are not tagged EFI_MEMORY_RUNTIME, since those
494 * regions have now been freed.
495 */
496 new_md = new;
497 for_each_efi_memory_desc(md) {
498 if (!(md->attribute & EFI_MEMORY_RUNTIME) &&
499 (md->type == EFI_BOOT_SERVICES_CODE ||
500 md->type == EFI_BOOT_SERVICES_DATA))
501 continue;
502
503 memcpy(new_md, md, efi.memmap.desc_size);
504 new_md += efi.memmap.desc_size;
505 }
506
507 memunmap(new);
508
509 if (efi_memmap_install(&data) != 0) {
510 pr_err("Could not install new EFI memmap\n");
511 return;
512 }
513 }
514
515 /*
516 * A number of config table entries get remapped to virtual addresses
517 * after entering EFI virtual mode. However, the kexec kernel requires
518 * their physical addresses therefore we pass them via setup_data and
519 * correct those entries to their respective physical addresses here.
520 *
521 * Currently only handles smbios which is necessary for some firmware
522 * implementation.
523 */
efi_reuse_config(u64 tables,int nr_tables)524 int __init efi_reuse_config(u64 tables, int nr_tables)
525 {
526 int i, sz, ret = 0;
527 void *p, *tablep;
528 struct efi_setup_data *data;
529
530 if (nr_tables == 0)
531 return 0;
532
533 if (!efi_setup)
534 return 0;
535
536 if (!efi_enabled(EFI_64BIT))
537 return 0;
538
539 data = early_memremap(efi_setup, sizeof(*data));
540 if (!data) {
541 ret = -ENOMEM;
542 goto out;
543 }
544
545 if (!data->smbios)
546 goto out_memremap;
547
548 sz = sizeof(efi_config_table_64_t);
549
550 p = tablep = early_memremap(tables, nr_tables * sz);
551 if (!p) {
552 pr_err("Could not map Configuration table!\n");
553 ret = -ENOMEM;
554 goto out_memremap;
555 }
556
557 for (i = 0; i < nr_tables; i++) {
558 efi_guid_t guid;
559
560 guid = ((efi_config_table_64_t *)p)->guid;
561
562 if (!efi_guidcmp(guid, SMBIOS_TABLE_GUID))
563 ((efi_config_table_64_t *)p)->table = data->smbios;
564 p += sz;
565 }
566 early_memunmap(tablep, nr_tables * sz);
567
568 out_memremap:
569 early_memunmap(data, sizeof(*data));
570 out:
571 return ret;
572 }
573
efi_apply_memmap_quirks(void)574 void __init efi_apply_memmap_quirks(void)
575 {
576 /*
577 * Once setup is done earlier, unmap the EFI memory map on mismatched
578 * firmware/kernel architectures since there is no support for runtime
579 * services.
580 */
581 if (!efi_runtime_supported()) {
582 pr_info("Setup done, disabling due to 32/64-bit mismatch\n");
583 efi_memmap_unmap();
584 }
585 }
586
587 /*
588 * For most modern platforms the preferred method of powering off is via
589 * ACPI. However, there are some that are known to require the use of
590 * EFI runtime services and for which ACPI does not work at all.
591 *
592 * Using EFI is a last resort, to be used only if no other option
593 * exists.
594 */
efi_reboot_required(void)595 bool efi_reboot_required(void)
596 {
597 if (!acpi_gbl_reduced_hardware)
598 return false;
599
600 efi_reboot_quirk_mode = EFI_RESET_WARM;
601 return true;
602 }
603
efi_poweroff_required(void)604 bool efi_poweroff_required(void)
605 {
606 return acpi_gbl_reduced_hardware || acpi_no_s5;
607 }
608
609 #ifdef CONFIG_EFI_CAPSULE_QUIRK_QUARK_CSH
610
qrk_capsule_setup_info(struct capsule_info * cap_info,void ** pkbuff,size_t hdr_bytes)611 static int qrk_capsule_setup_info(struct capsule_info *cap_info, void **pkbuff,
612 size_t hdr_bytes)
613 {
614 struct quark_security_header *csh = *pkbuff;
615
616 /* Only process data block that is larger than the security header */
617 if (hdr_bytes < sizeof(struct quark_security_header))
618 return 0;
619
620 if (csh->csh_signature != QUARK_CSH_SIGNATURE ||
621 csh->headersize != QUARK_SECURITY_HEADER_SIZE)
622 return 1;
623
624 /* Only process data block if EFI header is included */
625 if (hdr_bytes < QUARK_SECURITY_HEADER_SIZE +
626 sizeof(efi_capsule_header_t))
627 return 0;
628
629 pr_debug("Quark security header detected\n");
630
631 if (csh->rsvd_next_header != 0) {
632 pr_err("multiple Quark security headers not supported\n");
633 return -EINVAL;
634 }
635
636 *pkbuff += csh->headersize;
637 cap_info->total_size = csh->headersize;
638
639 /*
640 * Update the first page pointer to skip over the CSH header.
641 */
642 cap_info->phys[0] += csh->headersize;
643
644 /*
645 * cap_info->capsule should point at a virtual mapping of the entire
646 * capsule, starting at the capsule header. Our image has the Quark
647 * security header prepended, so we cannot rely on the default vmap()
648 * mapping created by the generic capsule code.
649 * Given that the Quark firmware does not appear to care about the
650 * virtual mapping, let's just point cap_info->capsule at our copy
651 * of the capsule header.
652 */
653 cap_info->capsule = &cap_info->header;
654
655 return 1;
656 }
657
658 static const struct x86_cpu_id efi_capsule_quirk_ids[] = {
659 X86_MATCH_VENDOR_FAM_MODEL(INTEL, 5, INTEL_FAM5_QUARK_X1000,
660 &qrk_capsule_setup_info),
661 { }
662 };
663
efi_capsule_setup_info(struct capsule_info * cap_info,void * kbuff,size_t hdr_bytes)664 int efi_capsule_setup_info(struct capsule_info *cap_info, void *kbuff,
665 size_t hdr_bytes)
666 {
667 int (*quirk_handler)(struct capsule_info *, void **, size_t);
668 const struct x86_cpu_id *id;
669 int ret;
670
671 if (hdr_bytes < sizeof(efi_capsule_header_t))
672 return 0;
673
674 cap_info->total_size = 0;
675
676 id = x86_match_cpu(efi_capsule_quirk_ids);
677 if (id) {
678 /*
679 * The quirk handler is supposed to return
680 * - a value > 0 if the setup should continue, after advancing
681 * kbuff as needed
682 * - 0 if not enough hdr_bytes are available yet
683 * - a negative error code otherwise
684 */
685 quirk_handler = (typeof(quirk_handler))id->driver_data;
686 ret = quirk_handler(cap_info, &kbuff, hdr_bytes);
687 if (ret <= 0)
688 return ret;
689 }
690
691 memcpy(&cap_info->header, kbuff, sizeof(cap_info->header));
692
693 cap_info->total_size += cap_info->header.imagesize;
694
695 return __efi_capsule_setup_info(cap_info);
696 }
697
698 #endif
699
700 /*
701 * If any access by any efi runtime service causes a page fault, then,
702 * 1. If it's efi_reset_system(), reboot through BIOS.
703 * 2. If any other efi runtime service, then
704 * a. Return error status to the efi caller process.
705 * b. Disable EFI Runtime Services forever and
706 * c. Freeze efi_rts_wq and schedule new process.
707 *
708 * @return: Returns, if the page fault is not handled. This function
709 * will never return if the page fault is handled successfully.
710 */
efi_crash_gracefully_on_page_fault(unsigned long phys_addr)711 void efi_crash_gracefully_on_page_fault(unsigned long phys_addr)
712 {
713 if (!IS_ENABLED(CONFIG_X86_64))
714 return;
715
716 /*
717 * If we get an interrupt/NMI while processing an EFI runtime service
718 * then this is a regular OOPS, not an EFI failure.
719 */
720 if (in_interrupt())
721 return;
722
723 /*
724 * Make sure that an efi runtime service caused the page fault.
725 * READ_ONCE() because we might be OOPSing in a different thread,
726 * and we don't want to trip KTSAN while trying to OOPS.
727 */
728 if (READ_ONCE(efi_rts_work.efi_rts_id) == EFI_NONE ||
729 current_work() != &efi_rts_work.work)
730 return;
731
732 /*
733 * Address range 0x0000 - 0x0fff is always mapped in the efi_pgd, so
734 * page faulting on these addresses isn't expected.
735 */
736 if (phys_addr <= 0x0fff)
737 return;
738
739 /*
740 * Print stack trace as it might be useful to know which EFI Runtime
741 * Service is buggy.
742 */
743 WARN(1, FW_BUG "Page fault caused by firmware at PA: 0x%lx\n",
744 phys_addr);
745
746 /*
747 * Buggy efi_reset_system() is handled differently from other EFI
748 * Runtime Services as it doesn't use efi_rts_wq. Although,
749 * native_machine_emergency_restart() says that machine_real_restart()
750 * could fail, it's better not to complicate this fault handler
751 * because this case occurs *very* rarely and hence could be improved
752 * on a need by basis.
753 */
754 if (efi_rts_work.efi_rts_id == EFI_RESET_SYSTEM) {
755 pr_info("efi_reset_system() buggy! Reboot through BIOS\n");
756 machine_real_restart(MRR_BIOS);
757 return;
758 }
759
760 /*
761 * Before calling EFI Runtime Service, the kernel has switched the
762 * calling process to efi_mm. Hence, switch back to task_mm.
763 */
764 arch_efi_call_virt_teardown();
765
766 /* Signal error status to the efi caller process */
767 efi_rts_work.status = EFI_ABORTED;
768 complete(&efi_rts_work.efi_rts_comp);
769
770 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
771 pr_info("Froze efi_rts_wq and disabled EFI Runtime Services\n");
772
773 /*
774 * Call schedule() in an infinite loop, so that any spurious wake ups
775 * will never run efi_rts_wq again.
776 */
777 for (;;) {
778 set_current_state(TASK_IDLE);
779 schedule();
780 }
781 }
782