xref: /openbmc/linux/arch/x86/platform/efi/quirks.c (revision 532c2919)
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 	if (!efi_enabled(EFI_MEMMAP))
322 		return;
323 
324 	for_each_efi_memory_desc(md) {
325 		u64 start = md->phys_addr;
326 		u64 size = md->num_pages << EFI_PAGE_SHIFT;
327 		bool already_reserved;
328 
329 		if (md->type != EFI_BOOT_SERVICES_CODE &&
330 		    md->type != EFI_BOOT_SERVICES_DATA)
331 			continue;
332 
333 		already_reserved = memblock_is_region_reserved(start, size);
334 
335 		/*
336 		 * Because the following memblock_reserve() is paired
337 		 * with memblock_free_late() for this region in
338 		 * efi_free_boot_services(), we must be extremely
339 		 * careful not to reserve, and subsequently free,
340 		 * critical regions of memory (like the kernel image) or
341 		 * those regions that somebody else has already
342 		 * reserved.
343 		 *
344 		 * A good example of a critical region that must not be
345 		 * freed is page zero (first 4Kb of memory), which may
346 		 * contain boot services code/data but is marked
347 		 * E820_TYPE_RESERVED by trim_bios_range().
348 		 */
349 		if (!already_reserved) {
350 			memblock_reserve(start, size);
351 
352 			/*
353 			 * If we are the first to reserve the region, no
354 			 * one else cares about it. We own it and can
355 			 * free it later.
356 			 */
357 			if (can_free_region(start, size))
358 				continue;
359 		}
360 
361 		/*
362 		 * We don't own the region. We must not free it.
363 		 *
364 		 * Setting this bit for a boot services region really
365 		 * doesn't make sense as far as the firmware is
366 		 * concerned, but it does provide us with a way to tag
367 		 * those regions that must not be paired with
368 		 * memblock_free_late().
369 		 */
370 		md->attribute |= EFI_MEMORY_RUNTIME;
371 	}
372 }
373 
374 /*
375  * Apart from having VA mappings for EFI boot services code/data regions,
376  * (duplicate) 1:1 mappings were also created as a quirk for buggy firmware. So,
377  * unmap both 1:1 and VA mappings.
378  */
379 static void __init efi_unmap_pages(efi_memory_desc_t *md)
380 {
381 	pgd_t *pgd = efi_mm.pgd;
382 	u64 pa = md->phys_addr;
383 	u64 va = md->virt_addr;
384 
385 	/*
386 	 * To Do: Remove this check after adding functionality to unmap EFI boot
387 	 * services code/data regions from direct mapping area because
388 	 * "efi=old_map" maps EFI regions in swapper_pg_dir.
389 	 */
390 	if (efi_enabled(EFI_OLD_MEMMAP))
391 		return;
392 
393 	/*
394 	 * EFI mixed mode has all RAM mapped to access arguments while making
395 	 * EFI runtime calls, hence don't unmap EFI boot services code/data
396 	 * regions.
397 	 */
398 	if (!efi_is_native())
399 		return;
400 
401 	if (kernel_unmap_pages_in_pgd(pgd, pa, md->num_pages))
402 		pr_err("Failed to unmap 1:1 mapping for 0x%llx\n", pa);
403 
404 	if (kernel_unmap_pages_in_pgd(pgd, va, md->num_pages))
405 		pr_err("Failed to unmap VA mapping for 0x%llx\n", va);
406 }
407 
408 void __init efi_free_boot_services(void)
409 {
410 	phys_addr_t new_phys, new_size;
411 	efi_memory_desc_t *md;
412 	int num_entries = 0;
413 	void *new, *new_md;
414 
415 	for_each_efi_memory_desc(md) {
416 		unsigned long long start = md->phys_addr;
417 		unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
418 		size_t rm_size;
419 
420 		if (md->type != EFI_BOOT_SERVICES_CODE &&
421 		    md->type != EFI_BOOT_SERVICES_DATA) {
422 			num_entries++;
423 			continue;
424 		}
425 
426 		/* Do not free, someone else owns it: */
427 		if (md->attribute & EFI_MEMORY_RUNTIME) {
428 			num_entries++;
429 			continue;
430 		}
431 
432 		/*
433 		 * Before calling set_virtual_address_map(), EFI boot services
434 		 * code/data regions were mapped as a quirk for buggy firmware.
435 		 * Unmap them from efi_pgd before freeing them up.
436 		 */
437 		efi_unmap_pages(md);
438 
439 		/*
440 		 * Nasty quirk: if all sub-1MB memory is used for boot
441 		 * services, we can get here without having allocated the
442 		 * real mode trampoline.  It's too late to hand boot services
443 		 * memory back to the memblock allocator, so instead
444 		 * try to manually allocate the trampoline if needed.
445 		 *
446 		 * I've seen this on a Dell XPS 13 9350 with firmware
447 		 * 1.4.4 with SGX enabled booting Linux via Fedora 24's
448 		 * grub2-efi on a hard disk.  (And no, I don't know why
449 		 * this happened, but Linux should still try to boot rather
450 		 * panicing early.)
451 		 */
452 		rm_size = real_mode_size_needed();
453 		if (rm_size && (start + rm_size) < (1<<20) && size >= rm_size) {
454 			set_real_mode_mem(start);
455 			start += rm_size;
456 			size -= rm_size;
457 		}
458 
459 		memblock_free_late(start, size);
460 	}
461 
462 	if (!num_entries)
463 		return;
464 
465 	new_size = efi.memmap.desc_size * num_entries;
466 	new_phys = efi_memmap_alloc(num_entries);
467 	if (!new_phys) {
468 		pr_err("Failed to allocate new EFI memmap\n");
469 		return;
470 	}
471 
472 	new = memremap(new_phys, new_size, MEMREMAP_WB);
473 	if (!new) {
474 		pr_err("Failed to map new EFI memmap\n");
475 		return;
476 	}
477 
478 	/*
479 	 * Build a new EFI memmap that excludes any boot services
480 	 * regions that are not tagged EFI_MEMORY_RUNTIME, since those
481 	 * regions have now been freed.
482 	 */
483 	new_md = new;
484 	for_each_efi_memory_desc(md) {
485 		if (!(md->attribute & EFI_MEMORY_RUNTIME) &&
486 		    (md->type == EFI_BOOT_SERVICES_CODE ||
487 		     md->type == EFI_BOOT_SERVICES_DATA))
488 			continue;
489 
490 		memcpy(new_md, md, efi.memmap.desc_size);
491 		new_md += efi.memmap.desc_size;
492 	}
493 
494 	memunmap(new);
495 
496 	if (efi_memmap_install(new_phys, num_entries)) {
497 		pr_err("Could not install new EFI memmap\n");
498 		return;
499 	}
500 }
501 
502 /*
503  * A number of config table entries get remapped to virtual addresses
504  * after entering EFI virtual mode. However, the kexec kernel requires
505  * their physical addresses therefore we pass them via setup_data and
506  * correct those entries to their respective physical addresses here.
507  *
508  * Currently only handles smbios which is necessary for some firmware
509  * implementation.
510  */
511 int __init efi_reuse_config(u64 tables, int nr_tables)
512 {
513 	int i, sz, ret = 0;
514 	void *p, *tablep;
515 	struct efi_setup_data *data;
516 
517 	if (nr_tables == 0)
518 		return 0;
519 
520 	if (!efi_setup)
521 		return 0;
522 
523 	if (!efi_enabled(EFI_64BIT))
524 		return 0;
525 
526 	data = early_memremap(efi_setup, sizeof(*data));
527 	if (!data) {
528 		ret = -ENOMEM;
529 		goto out;
530 	}
531 
532 	if (!data->smbios)
533 		goto out_memremap;
534 
535 	sz = sizeof(efi_config_table_64_t);
536 
537 	p = tablep = early_memremap(tables, nr_tables * sz);
538 	if (!p) {
539 		pr_err("Could not map Configuration table!\n");
540 		ret = -ENOMEM;
541 		goto out_memremap;
542 	}
543 
544 	for (i = 0; i < efi.systab->nr_tables; i++) {
545 		efi_guid_t guid;
546 
547 		guid = ((efi_config_table_64_t *)p)->guid;
548 
549 		if (!efi_guidcmp(guid, SMBIOS_TABLE_GUID))
550 			((efi_config_table_64_t *)p)->table = data->smbios;
551 		p += sz;
552 	}
553 	early_memunmap(tablep, nr_tables * sz);
554 
555 out_memremap:
556 	early_memunmap(data, sizeof(*data));
557 out:
558 	return ret;
559 }
560 
561 static const struct dmi_system_id sgi_uv1_dmi[] = {
562 	{ NULL, "SGI UV1",
563 		{	DMI_MATCH(DMI_PRODUCT_NAME,	"Stoutland Platform"),
564 			DMI_MATCH(DMI_PRODUCT_VERSION,	"1.0"),
565 			DMI_MATCH(DMI_BIOS_VENDOR,	"SGI.COM"),
566 		}
567 	},
568 	{ } /* NULL entry stops DMI scanning */
569 };
570 
571 void __init efi_apply_memmap_quirks(void)
572 {
573 	/*
574 	 * Once setup is done earlier, unmap the EFI memory map on mismatched
575 	 * firmware/kernel architectures since there is no support for runtime
576 	 * services.
577 	 */
578 	if (!efi_runtime_supported()) {
579 		pr_info("Setup done, disabling due to 32/64-bit mismatch\n");
580 		efi_memmap_unmap();
581 	}
582 
583 	/* UV2+ BIOS has a fix for this issue.  UV1 still needs the quirk. */
584 	if (dmi_check_system(sgi_uv1_dmi))
585 		set_bit(EFI_OLD_MEMMAP, &efi.flags);
586 }
587 
588 /*
589  * For most modern platforms the preferred method of powering off is via
590  * ACPI. However, there are some that are known to require the use of
591  * EFI runtime services and for which ACPI does not work at all.
592  *
593  * Using EFI is a last resort, to be used only if no other option
594  * exists.
595  */
596 bool efi_reboot_required(void)
597 {
598 	if (!acpi_gbl_reduced_hardware)
599 		return false;
600 
601 	efi_reboot_quirk_mode = EFI_RESET_WARM;
602 	return true;
603 }
604 
605 bool efi_poweroff_required(void)
606 {
607 	return acpi_gbl_reduced_hardware || acpi_no_s5;
608 }
609 
610 #ifdef CONFIG_EFI_CAPSULE_QUIRK_QUARK_CSH
611 
612 static int qrk_capsule_setup_info(struct capsule_info *cap_info, void **pkbuff,
613 				  size_t hdr_bytes)
614 {
615 	struct quark_security_header *csh = *pkbuff;
616 
617 	/* Only process data block that is larger than the security header */
618 	if (hdr_bytes < sizeof(struct quark_security_header))
619 		return 0;
620 
621 	if (csh->csh_signature != QUARK_CSH_SIGNATURE ||
622 	    csh->headersize != QUARK_SECURITY_HEADER_SIZE)
623 		return 1;
624 
625 	/* Only process data block if EFI header is included */
626 	if (hdr_bytes < QUARK_SECURITY_HEADER_SIZE +
627 			sizeof(efi_capsule_header_t))
628 		return 0;
629 
630 	pr_debug("Quark security header detected\n");
631 
632 	if (csh->rsvd_next_header != 0) {
633 		pr_err("multiple Quark security headers not supported\n");
634 		return -EINVAL;
635 	}
636 
637 	*pkbuff += csh->headersize;
638 	cap_info->total_size = csh->headersize;
639 
640 	/*
641 	 * Update the first page pointer to skip over the CSH header.
642 	 */
643 	cap_info->phys[0] += csh->headersize;
644 
645 	/*
646 	 * cap_info->capsule should point at a virtual mapping of the entire
647 	 * capsule, starting at the capsule header. Our image has the Quark
648 	 * security header prepended, so we cannot rely on the default vmap()
649 	 * mapping created by the generic capsule code.
650 	 * Given that the Quark firmware does not appear to care about the
651 	 * virtual mapping, let's just point cap_info->capsule at our copy
652 	 * of the capsule header.
653 	 */
654 	cap_info->capsule = &cap_info->header;
655 
656 	return 1;
657 }
658 
659 #define ICPU(family, model, quirk_handler) \
660 	{ X86_VENDOR_INTEL, family, model, X86_FEATURE_ANY, \
661 	  (unsigned long)&quirk_handler }
662 
663 static const struct x86_cpu_id efi_capsule_quirk_ids[] = {
664 	ICPU(5, 9, qrk_capsule_setup_info),	/* Intel Quark X1000 */
665 	{ }
666 };
667 
668 int efi_capsule_setup_info(struct capsule_info *cap_info, void *kbuff,
669 			   size_t hdr_bytes)
670 {
671 	int (*quirk_handler)(struct capsule_info *, void **, size_t);
672 	const struct x86_cpu_id *id;
673 	int ret;
674 
675 	if (hdr_bytes < sizeof(efi_capsule_header_t))
676 		return 0;
677 
678 	cap_info->total_size = 0;
679 
680 	id = x86_match_cpu(efi_capsule_quirk_ids);
681 	if (id) {
682 		/*
683 		 * The quirk handler is supposed to return
684 		 *  - a value > 0 if the setup should continue, after advancing
685 		 *    kbuff as needed
686 		 *  - 0 if not enough hdr_bytes are available yet
687 		 *  - a negative error code otherwise
688 		 */
689 		quirk_handler = (typeof(quirk_handler))id->driver_data;
690 		ret = quirk_handler(cap_info, &kbuff, hdr_bytes);
691 		if (ret <= 0)
692 			return ret;
693 	}
694 
695 	memcpy(&cap_info->header, kbuff, sizeof(cap_info->header));
696 
697 	cap_info->total_size += cap_info->header.imagesize;
698 
699 	return __efi_capsule_setup_info(cap_info);
700 }
701 
702 #endif
703 
704 /*
705  * If any access by any efi runtime service causes a page fault, then,
706  * 1. If it's efi_reset_system(), reboot through BIOS.
707  * 2. If any other efi runtime service, then
708  *    a. Return error status to the efi caller process.
709  *    b. Disable EFI Runtime Services forever and
710  *    c. Freeze efi_rts_wq and schedule new process.
711  *
712  * @return: Returns, if the page fault is not handled. This function
713  * will never return if the page fault is handled successfully.
714  */
715 void efi_recover_from_page_fault(unsigned long phys_addr)
716 {
717 	if (!IS_ENABLED(CONFIG_X86_64))
718 		return;
719 
720 	/*
721 	 * Make sure that an efi runtime service caused the page fault.
722 	 * "efi_mm" cannot be used to check if the page fault had occurred
723 	 * in the firmware context because efi=old_map doesn't use efi_pgd.
724 	 */
725 	if (efi_rts_work.efi_rts_id == EFI_NONE)
726 		return;
727 
728 	/*
729 	 * Address range 0x0000 - 0x0fff is always mapped in the efi_pgd, so
730 	 * page faulting on these addresses isn't expected.
731 	 */
732 	if (phys_addr <= 0x0fff)
733 		return;
734 
735 	/*
736 	 * Print stack trace as it might be useful to know which EFI Runtime
737 	 * Service is buggy.
738 	 */
739 	WARN(1, FW_BUG "Page fault caused by firmware at PA: 0x%lx\n",
740 	     phys_addr);
741 
742 	/*
743 	 * Buggy efi_reset_system() is handled differently from other EFI
744 	 * Runtime Services as it doesn't use efi_rts_wq. Although,
745 	 * native_machine_emergency_restart() says that machine_real_restart()
746 	 * could fail, it's better not to compilcate this fault handler
747 	 * because this case occurs *very* rarely and hence could be improved
748 	 * on a need by basis.
749 	 */
750 	if (efi_rts_work.efi_rts_id == EFI_RESET_SYSTEM) {
751 		pr_info("efi_reset_system() buggy! Reboot through BIOS\n");
752 		machine_real_restart(MRR_BIOS);
753 		return;
754 	}
755 
756 	/*
757 	 * Before calling EFI Runtime Service, the kernel has switched the
758 	 * calling process to efi_mm. Hence, switch back to task_mm.
759 	 */
760 	arch_efi_call_virt_teardown();
761 
762 	/* Signal error status to the efi caller process */
763 	efi_rts_work.status = EFI_ABORTED;
764 	complete(&efi_rts_work.efi_rts_comp);
765 
766 	clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
767 	pr_info("Froze efi_rts_wq and disabled EFI Runtime Services\n");
768 
769 	/*
770 	 * Call schedule() in an infinite loop, so that any spurious wake ups
771 	 * will never run efi_rts_wq again.
772 	 */
773 	for (;;) {
774 		set_current_state(TASK_IDLE);
775 		schedule();
776 	}
777 
778 	return;
779 }
780