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
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
query_variable_store_nonblocking(u32 attributes,unsigned long size)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 */
efi_query_variable_store(u32 attributes,unsigned long size,bool nonblocking)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 */
efi_arch_mem_reserve(phys_addr_t addr,u64 size)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_prot(data.phys_map, data.size,
281 pgprot_val(pgprot_encrypted(FIXMAP_PAGE_NORMAL)));
282 if (!new) {
283 pr_err("Failed to map new boot services memmap\n");
284 return;
285 }
286
287 efi_memmap_insert(&efi.memmap, new, &mr);
288 early_memunmap(new, data.size);
289
290 efi_memmap_install(&data);
291 e820__range_update(addr, size, E820_TYPE_RAM, E820_TYPE_RESERVED);
292 e820__update_table(e820_table);
293 }
294
295 /*
296 * Helper function for efi_reserve_boot_services() to figure out if we
297 * can free regions in efi_free_boot_services().
298 *
299 * Use this function to ensure we do not free regions owned by somebody
300 * else. We must only reserve (and then free) regions:
301 *
302 * - Not within any part of the kernel
303 * - Not the BIOS reserved area (E820_TYPE_RESERVED, E820_TYPE_NVS, etc)
304 */
can_free_region(u64 start,u64 size)305 static __init bool can_free_region(u64 start, u64 size)
306 {
307 if (start + size > __pa_symbol(_text) && start <= __pa_symbol(_end))
308 return false;
309
310 if (!e820__mapped_all(start, start+size, E820_TYPE_RAM))
311 return false;
312
313 return true;
314 }
315
efi_reserve_boot_services(void)316 void __init efi_reserve_boot_services(void)
317 {
318 efi_memory_desc_t *md;
319
320 if (!efi_enabled(EFI_MEMMAP))
321 return;
322
323 for_each_efi_memory_desc(md) {
324 u64 start = md->phys_addr;
325 u64 size = md->num_pages << EFI_PAGE_SHIFT;
326 bool already_reserved;
327
328 if (md->type != EFI_BOOT_SERVICES_CODE &&
329 md->type != EFI_BOOT_SERVICES_DATA)
330 continue;
331
332 already_reserved = memblock_is_region_reserved(start, size);
333
334 /*
335 * Because the following memblock_reserve() is paired
336 * with memblock_free_late() for this region in
337 * efi_free_boot_services(), we must be extremely
338 * careful not to reserve, and subsequently free,
339 * critical regions of memory (like the kernel image) or
340 * those regions that somebody else has already
341 * reserved.
342 *
343 * A good example of a critical region that must not be
344 * freed is page zero (first 4Kb of memory), which may
345 * contain boot services code/data but is marked
346 * E820_TYPE_RESERVED by trim_bios_range().
347 */
348 if (!already_reserved) {
349 memblock_reserve(start, size);
350
351 /*
352 * If we are the first to reserve the region, no
353 * one else cares about it. We own it and can
354 * free it later.
355 */
356 if (can_free_region(start, size))
357 continue;
358 }
359
360 /*
361 * We don't own the region. We must not free it.
362 *
363 * Setting this bit for a boot services region really
364 * doesn't make sense as far as the firmware is
365 * concerned, but it does provide us with a way to tag
366 * those regions that must not be paired with
367 * memblock_free_late().
368 */
369 md->attribute |= EFI_MEMORY_RUNTIME;
370 }
371 }
372
373 /*
374 * Apart from having VA mappings for EFI boot services code/data regions,
375 * (duplicate) 1:1 mappings were also created as a quirk for buggy firmware. So,
376 * unmap both 1:1 and VA mappings.
377 */
efi_unmap_pages(efi_memory_desc_t * md)378 static void __init efi_unmap_pages(efi_memory_desc_t *md)
379 {
380 pgd_t *pgd = efi_mm.pgd;
381 u64 pa = md->phys_addr;
382 u64 va = md->virt_addr;
383
384 /*
385 * EFI mixed mode has all RAM mapped to access arguments while making
386 * EFI runtime calls, hence don't unmap EFI boot services code/data
387 * regions.
388 */
389 if (efi_is_mixed())
390 return;
391
392 if (kernel_unmap_pages_in_pgd(pgd, pa, md->num_pages))
393 pr_err("Failed to unmap 1:1 mapping for 0x%llx\n", pa);
394
395 if (kernel_unmap_pages_in_pgd(pgd, va, md->num_pages))
396 pr_err("Failed to unmap VA mapping for 0x%llx\n", va);
397 }
398
efi_free_boot_services(void)399 void __init efi_free_boot_services(void)
400 {
401 struct efi_memory_map_data data = { 0 };
402 efi_memory_desc_t *md;
403 int num_entries = 0;
404 void *new, *new_md;
405
406 /* Keep all regions for /sys/kernel/debug/efi */
407 if (efi_enabled(EFI_DBG))
408 return;
409
410 for_each_efi_memory_desc(md) {
411 unsigned long long start = md->phys_addr;
412 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
413 size_t rm_size;
414
415 if (md->type != EFI_BOOT_SERVICES_CODE &&
416 md->type != EFI_BOOT_SERVICES_DATA) {
417 num_entries++;
418 continue;
419 }
420
421 /* Do not free, someone else owns it: */
422 if (md->attribute & EFI_MEMORY_RUNTIME) {
423 num_entries++;
424 continue;
425 }
426
427 /*
428 * Before calling set_virtual_address_map(), EFI boot services
429 * code/data regions were mapped as a quirk for buggy firmware.
430 * Unmap them from efi_pgd before freeing them up.
431 */
432 efi_unmap_pages(md);
433
434 /*
435 * Nasty quirk: if all sub-1MB memory is used for boot
436 * services, we can get here without having allocated the
437 * real mode trampoline. It's too late to hand boot services
438 * memory back to the memblock allocator, so instead
439 * try to manually allocate the trampoline if needed.
440 *
441 * I've seen this on a Dell XPS 13 9350 with firmware
442 * 1.4.4 with SGX enabled booting Linux via Fedora 24's
443 * grub2-efi on a hard disk. (And no, I don't know why
444 * this happened, but Linux should still try to boot rather
445 * panicking early.)
446 */
447 rm_size = real_mode_size_needed();
448 if (rm_size && (start + rm_size) < (1<<20) && size >= rm_size) {
449 set_real_mode_mem(start);
450 start += rm_size;
451 size -= rm_size;
452 }
453
454 /*
455 * Don't free memory under 1M for two reasons:
456 * - BIOS might clobber it
457 * - Crash kernel needs it to be reserved
458 */
459 if (start + size < SZ_1M)
460 continue;
461 if (start < SZ_1M) {
462 size -= (SZ_1M - start);
463 start = SZ_1M;
464 }
465
466 memblock_free_late(start, size);
467 }
468
469 if (!num_entries)
470 return;
471
472 if (efi_memmap_alloc(num_entries, &data) != 0) {
473 pr_err("Failed to allocate new EFI memmap\n");
474 return;
475 }
476
477 new = memremap(data.phys_map, data.size, MEMREMAP_WB);
478 if (!new) {
479 pr_err("Failed to map new EFI memmap\n");
480 return;
481 }
482
483 /*
484 * Build a new EFI memmap that excludes any boot services
485 * regions that are not tagged EFI_MEMORY_RUNTIME, since those
486 * regions have now been freed.
487 */
488 new_md = new;
489 for_each_efi_memory_desc(md) {
490 if (!(md->attribute & EFI_MEMORY_RUNTIME) &&
491 (md->type == EFI_BOOT_SERVICES_CODE ||
492 md->type == EFI_BOOT_SERVICES_DATA))
493 continue;
494
495 memcpy(new_md, md, efi.memmap.desc_size);
496 new_md += efi.memmap.desc_size;
497 }
498
499 memunmap(new);
500
501 if (efi_memmap_install(&data) != 0) {
502 pr_err("Could not install new EFI memmap\n");
503 return;
504 }
505 }
506
507 /*
508 * A number of config table entries get remapped to virtual addresses
509 * after entering EFI virtual mode. However, the kexec kernel requires
510 * their physical addresses therefore we pass them via setup_data and
511 * correct those entries to their respective physical addresses here.
512 *
513 * Currently only handles smbios which is necessary for some firmware
514 * implementation.
515 */
efi_reuse_config(u64 tables,int nr_tables)516 int __init efi_reuse_config(u64 tables, int nr_tables)
517 {
518 int i, sz, ret = 0;
519 void *p, *tablep;
520 struct efi_setup_data *data;
521
522 if (nr_tables == 0)
523 return 0;
524
525 if (!efi_setup)
526 return 0;
527
528 if (!efi_enabled(EFI_64BIT))
529 return 0;
530
531 data = early_memremap(efi_setup, sizeof(*data));
532 if (!data) {
533 ret = -ENOMEM;
534 goto out;
535 }
536
537 if (!data->smbios)
538 goto out_memremap;
539
540 sz = sizeof(efi_config_table_64_t);
541
542 p = tablep = early_memremap(tables, nr_tables * sz);
543 if (!p) {
544 pr_err("Could not map Configuration table!\n");
545 ret = -ENOMEM;
546 goto out_memremap;
547 }
548
549 for (i = 0; i < nr_tables; i++) {
550 efi_guid_t guid;
551
552 guid = ((efi_config_table_64_t *)p)->guid;
553
554 if (!efi_guidcmp(guid, SMBIOS_TABLE_GUID))
555 ((efi_config_table_64_t *)p)->table = data->smbios;
556 p += sz;
557 }
558 early_memunmap(tablep, nr_tables * sz);
559
560 out_memremap:
561 early_memunmap(data, sizeof(*data));
562 out:
563 return ret;
564 }
565
efi_apply_memmap_quirks(void)566 void __init efi_apply_memmap_quirks(void)
567 {
568 /*
569 * Once setup is done earlier, unmap the EFI memory map on mismatched
570 * firmware/kernel architectures since there is no support for runtime
571 * services.
572 */
573 if (!efi_runtime_supported()) {
574 pr_info("Setup done, disabling due to 32/64-bit mismatch\n");
575 efi_memmap_unmap();
576 }
577 }
578
579 /*
580 * For most modern platforms the preferred method of powering off is via
581 * ACPI. However, there are some that are known to require the use of
582 * EFI runtime services and for which ACPI does not work at all.
583 *
584 * Using EFI is a last resort, to be used only if no other option
585 * exists.
586 */
efi_reboot_required(void)587 bool efi_reboot_required(void)
588 {
589 if (!acpi_gbl_reduced_hardware)
590 return false;
591
592 efi_reboot_quirk_mode = EFI_RESET_WARM;
593 return true;
594 }
595
efi_poweroff_required(void)596 bool efi_poweroff_required(void)
597 {
598 return acpi_gbl_reduced_hardware || acpi_no_s5;
599 }
600
601 #ifdef CONFIG_EFI_CAPSULE_QUIRK_QUARK_CSH
602
qrk_capsule_setup_info(struct capsule_info * cap_info,void ** pkbuff,size_t hdr_bytes)603 static int qrk_capsule_setup_info(struct capsule_info *cap_info, void **pkbuff,
604 size_t hdr_bytes)
605 {
606 struct quark_security_header *csh = *pkbuff;
607
608 /* Only process data block that is larger than the security header */
609 if (hdr_bytes < sizeof(struct quark_security_header))
610 return 0;
611
612 if (csh->csh_signature != QUARK_CSH_SIGNATURE ||
613 csh->headersize != QUARK_SECURITY_HEADER_SIZE)
614 return 1;
615
616 /* Only process data block if EFI header is included */
617 if (hdr_bytes < QUARK_SECURITY_HEADER_SIZE +
618 sizeof(efi_capsule_header_t))
619 return 0;
620
621 pr_debug("Quark security header detected\n");
622
623 if (csh->rsvd_next_header != 0) {
624 pr_err("multiple Quark security headers not supported\n");
625 return -EINVAL;
626 }
627
628 *pkbuff += csh->headersize;
629 cap_info->total_size = csh->headersize;
630
631 /*
632 * Update the first page pointer to skip over the CSH header.
633 */
634 cap_info->phys[0] += csh->headersize;
635
636 /*
637 * cap_info->capsule should point at a virtual mapping of the entire
638 * capsule, starting at the capsule header. Our image has the Quark
639 * security header prepended, so we cannot rely on the default vmap()
640 * mapping created by the generic capsule code.
641 * Given that the Quark firmware does not appear to care about the
642 * virtual mapping, let's just point cap_info->capsule at our copy
643 * of the capsule header.
644 */
645 cap_info->capsule = &cap_info->header;
646
647 return 1;
648 }
649
650 static const struct x86_cpu_id efi_capsule_quirk_ids[] = {
651 X86_MATCH_VENDOR_FAM_MODEL(INTEL, 5, INTEL_FAM5_QUARK_X1000,
652 &qrk_capsule_setup_info),
653 { }
654 };
655
efi_capsule_setup_info(struct capsule_info * cap_info,void * kbuff,size_t hdr_bytes)656 int efi_capsule_setup_info(struct capsule_info *cap_info, void *kbuff,
657 size_t hdr_bytes)
658 {
659 int (*quirk_handler)(struct capsule_info *, void **, size_t);
660 const struct x86_cpu_id *id;
661 int ret;
662
663 if (hdr_bytes < sizeof(efi_capsule_header_t))
664 return 0;
665
666 cap_info->total_size = 0;
667
668 id = x86_match_cpu(efi_capsule_quirk_ids);
669 if (id) {
670 /*
671 * The quirk handler is supposed to return
672 * - a value > 0 if the setup should continue, after advancing
673 * kbuff as needed
674 * - 0 if not enough hdr_bytes are available yet
675 * - a negative error code otherwise
676 */
677 quirk_handler = (typeof(quirk_handler))id->driver_data;
678 ret = quirk_handler(cap_info, &kbuff, hdr_bytes);
679 if (ret <= 0)
680 return ret;
681 }
682
683 memcpy(&cap_info->header, kbuff, sizeof(cap_info->header));
684
685 cap_info->total_size += cap_info->header.imagesize;
686
687 return __efi_capsule_setup_info(cap_info);
688 }
689
690 #endif
691
692 /*
693 * If any access by any efi runtime service causes a page fault, then,
694 * 1. If it's efi_reset_system(), reboot through BIOS.
695 * 2. If any other efi runtime service, then
696 * a. Return error status to the efi caller process.
697 * b. Disable EFI Runtime Services forever and
698 * c. Freeze efi_rts_wq and schedule new process.
699 *
700 * @return: Returns, if the page fault is not handled. This function
701 * will never return if the page fault is handled successfully.
702 */
efi_crash_gracefully_on_page_fault(unsigned long phys_addr)703 void efi_crash_gracefully_on_page_fault(unsigned long phys_addr)
704 {
705 if (!IS_ENABLED(CONFIG_X86_64))
706 return;
707
708 /*
709 * If we get an interrupt/NMI while processing an EFI runtime service
710 * then this is a regular OOPS, not an EFI failure.
711 */
712 if (in_interrupt())
713 return;
714
715 /*
716 * Make sure that an efi runtime service caused the page fault.
717 * READ_ONCE() because we might be OOPSing in a different thread,
718 * and we don't want to trip KTSAN while trying to OOPS.
719 */
720 if (READ_ONCE(efi_rts_work.efi_rts_id) == EFI_NONE ||
721 current_work() != &efi_rts_work.work)
722 return;
723
724 /*
725 * Address range 0x0000 - 0x0fff is always mapped in the efi_pgd, so
726 * page faulting on these addresses isn't expected.
727 */
728 if (phys_addr <= 0x0fff)
729 return;
730
731 /*
732 * Print stack trace as it might be useful to know which EFI Runtime
733 * Service is buggy.
734 */
735 WARN(1, FW_BUG "Page fault caused by firmware at PA: 0x%lx\n",
736 phys_addr);
737
738 /*
739 * Buggy efi_reset_system() is handled differently from other EFI
740 * Runtime Services as it doesn't use efi_rts_wq. Although,
741 * native_machine_emergency_restart() says that machine_real_restart()
742 * could fail, it's better not to complicate this fault handler
743 * because this case occurs *very* rarely and hence could be improved
744 * on a need by basis.
745 */
746 if (efi_rts_work.efi_rts_id == EFI_RESET_SYSTEM) {
747 pr_info("efi_reset_system() buggy! Reboot through BIOS\n");
748 machine_real_restart(MRR_BIOS);
749 return;
750 }
751
752 /*
753 * Before calling EFI Runtime Service, the kernel has switched the
754 * calling process to efi_mm. Hence, switch back to task_mm.
755 */
756 arch_efi_call_virt_teardown();
757
758 /* Signal error status to the efi caller process */
759 efi_rts_work.status = EFI_ABORTED;
760 complete(&efi_rts_work.efi_rts_comp);
761
762 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
763 pr_info("Froze efi_rts_wq and disabled EFI Runtime Services\n");
764
765 /*
766 * Call schedule() in an infinite loop, so that any spurious wake ups
767 * will never run efi_rts_wq again.
768 */
769 for (;;) {
770 set_current_state(TASK_IDLE);
771 schedule();
772 }
773 }
774