xref: /linux/arch/x86/platform/efi/efi_64.c (revision d27656d02d85078c63f060fca9c5d99794791a75)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * x86_64 specific EFI support functions
4  * Based on Extensible Firmware Interface Specification version 1.0
5  *
6  * Copyright (C) 2005-2008 Intel Co.
7  *	Fenghua Yu <fenghua.yu@intel.com>
8  *	Bibo Mao <bibo.mao@intel.com>
9  *	Chandramouli Narayanan <mouli@linux.intel.com>
10  *	Huang Ying <ying.huang@intel.com>
11  *
12  * Code to convert EFI to E820 map has been implemented in elilo bootloader
13  * based on a EFI patch by Edgar Hucek. Based on the E820 map, the page table
14  * is setup appropriately for EFI runtime code.
15  * - mouli 06/14/2007.
16  *
17  */
18 
19 #define pr_fmt(fmt) "efi: " fmt
20 
21 #include <linux/kernel.h>
22 #include <linux/init.h>
23 #include <linux/mm.h>
24 #include <linux/types.h>
25 #include <linux/spinlock.h>
26 #include <linux/memblock.h>
27 #include <linux/ioport.h>
28 #include <linux/mc146818rtc.h>
29 #include <linux/efi.h>
30 #include <linux/export.h>
31 #include <linux/uaccess.h>
32 #include <linux/io.h>
33 #include <linux/reboot.h>
34 #include <linux/slab.h>
35 #include <linux/ucs2_string.h>
36 #include <linux/cc_platform.h>
37 #include <linux/sched/task.h>
38 
39 #include <asm/setup.h>
40 #include <asm/page.h>
41 #include <asm/e820/api.h>
42 #include <asm/tlbflush.h>
43 #include <asm/proto.h>
44 #include <asm/efi.h>
45 #include <asm/cacheflush.h>
46 #include <asm/fixmap.h>
47 #include <asm/realmode.h>
48 #include <asm/time.h>
49 #include <asm/pgalloc.h>
50 #include <asm/sev.h>
51 
52 /*
53  * We allocate runtime services regions top-down, starting from -4G, i.e.
54  * 0xffff_ffff_0000_0000 and limit EFI VA mapping space to 64G.
55  */
56 static u64 efi_va = EFI_VA_START;
57 static struct mm_struct *efi_prev_mm;
58 
59 /*
60  * We need our own copy of the higher levels of the page tables
61  * because we want to avoid inserting EFI region mappings (EFI_VA_END
62  * to EFI_VA_START) into the standard kernel page tables. Everything
63  * else can be shared, see efi_sync_low_kernel_mappings().
64  *
65  * We don't want the pgd on the pgd_list and cannot use pgd_alloc() for the
66  * allocation.
67  */
68 int __init efi_alloc_page_tables(void)
69 {
70 	pgd_t *pgd, *efi_pgd;
71 	p4d_t *p4d;
72 	pud_t *pud;
73 	gfp_t gfp_mask;
74 
75 	gfp_mask = GFP_KERNEL | __GFP_ZERO;
76 	efi_pgd = (pgd_t *)__get_free_pages(gfp_mask, PGD_ALLOCATION_ORDER);
77 	if (!efi_pgd)
78 		goto fail;
79 
80 	pgd = efi_pgd + pgd_index(EFI_VA_END);
81 	p4d = p4d_alloc(&init_mm, pgd, EFI_VA_END);
82 	if (!p4d)
83 		goto free_pgd;
84 
85 	pud = pud_alloc(&init_mm, p4d, EFI_VA_END);
86 	if (!pud)
87 		goto free_p4d;
88 
89 	efi_mm.pgd = efi_pgd;
90 	mm_init_cpumask(&efi_mm);
91 	init_new_context(NULL, &efi_mm);
92 
93 	return 0;
94 
95 free_p4d:
96 	if (pgtable_l5_enabled())
97 		free_page((unsigned long)pgd_page_vaddr(*pgd));
98 free_pgd:
99 	free_pages((unsigned long)efi_pgd, PGD_ALLOCATION_ORDER);
100 fail:
101 	return -ENOMEM;
102 }
103 
104 /*
105  * Add low kernel mappings for passing arguments to EFI functions.
106  */
107 void efi_sync_low_kernel_mappings(void)
108 {
109 	unsigned num_entries;
110 	pgd_t *pgd_k, *pgd_efi;
111 	p4d_t *p4d_k, *p4d_efi;
112 	pud_t *pud_k, *pud_efi;
113 	pgd_t *efi_pgd = efi_mm.pgd;
114 
115 	pgd_efi = efi_pgd + pgd_index(PAGE_OFFSET);
116 	pgd_k = pgd_offset_k(PAGE_OFFSET);
117 
118 	num_entries = pgd_index(EFI_VA_END) - pgd_index(PAGE_OFFSET);
119 	memcpy(pgd_efi, pgd_k, sizeof(pgd_t) * num_entries);
120 
121 	pgd_efi = efi_pgd + pgd_index(EFI_VA_END);
122 	pgd_k = pgd_offset_k(EFI_VA_END);
123 	p4d_efi = p4d_offset(pgd_efi, 0);
124 	p4d_k = p4d_offset(pgd_k, 0);
125 
126 	num_entries = p4d_index(EFI_VA_END);
127 	memcpy(p4d_efi, p4d_k, sizeof(p4d_t) * num_entries);
128 
129 	/*
130 	 * We share all the PUD entries apart from those that map the
131 	 * EFI regions. Copy around them.
132 	 */
133 	BUILD_BUG_ON((EFI_VA_START & ~PUD_MASK) != 0);
134 	BUILD_BUG_ON((EFI_VA_END & ~PUD_MASK) != 0);
135 
136 	p4d_efi = p4d_offset(pgd_efi, EFI_VA_END);
137 	p4d_k = p4d_offset(pgd_k, EFI_VA_END);
138 	pud_efi = pud_offset(p4d_efi, 0);
139 	pud_k = pud_offset(p4d_k, 0);
140 
141 	num_entries = pud_index(EFI_VA_END);
142 	memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
143 
144 	pud_efi = pud_offset(p4d_efi, EFI_VA_START);
145 	pud_k = pud_offset(p4d_k, EFI_VA_START);
146 
147 	num_entries = PTRS_PER_PUD - pud_index(EFI_VA_START);
148 	memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
149 }
150 
151 /*
152  * Wrapper for slow_virt_to_phys() that handles NULL addresses.
153  */
154 static inline phys_addr_t
155 virt_to_phys_or_null_size(void *va, unsigned long size)
156 {
157 	phys_addr_t pa;
158 
159 	if (!va)
160 		return 0;
161 
162 	if (virt_addr_valid(va))
163 		return virt_to_phys(va);
164 
165 	pa = slow_virt_to_phys(va);
166 
167 	/* check if the object crosses a page boundary */
168 	if (WARN_ON((pa ^ (pa + size - 1)) & PAGE_MASK))
169 		return 0;
170 
171 	return pa;
172 }
173 
174 #define virt_to_phys_or_null(addr)				\
175 	virt_to_phys_or_null_size((addr), sizeof(*(addr)))
176 
177 int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)
178 {
179 	unsigned long pfn, text, pf, rodata;
180 	struct page *page;
181 	unsigned npages;
182 	pgd_t *pgd = efi_mm.pgd;
183 
184 	/*
185 	 * It can happen that the physical address of new_memmap lands in memory
186 	 * which is not mapped in the EFI page table. Therefore we need to go
187 	 * and ident-map those pages containing the map before calling
188 	 * phys_efi_set_virtual_address_map().
189 	 */
190 	pfn = pa_memmap >> PAGE_SHIFT;
191 	pf = _PAGE_NX | _PAGE_RW | _PAGE_ENC;
192 	if (kernel_map_pages_in_pgd(pgd, pfn, pa_memmap, num_pages, pf)) {
193 		pr_err("Error ident-mapping new memmap (0x%lx)!\n", pa_memmap);
194 		return 1;
195 	}
196 
197 	/*
198 	 * Certain firmware versions are way too sentimental and still believe
199 	 * they are exclusive and unquestionable owners of the first physical page,
200 	 * even though they explicitly mark it as EFI_CONVENTIONAL_MEMORY
201 	 * (but then write-access it later during SetVirtualAddressMap()).
202 	 *
203 	 * Create a 1:1 mapping for this page, to avoid triple faults during early
204 	 * boot with such firmware. We are free to hand this page to the BIOS,
205 	 * as trim_bios_range() will reserve the first page and isolate it away
206 	 * from memory allocators anyway.
207 	 */
208 	if (kernel_map_pages_in_pgd(pgd, 0x0, 0x0, 1, pf)) {
209 		pr_err("Failed to create 1:1 mapping for the first page!\n");
210 		return 1;
211 	}
212 
213 	/*
214 	 * When SEV-ES is active, the GHCB as set by the kernel will be used
215 	 * by firmware. Create a 1:1 unencrypted mapping for each GHCB.
216 	 */
217 	if (sev_es_efi_map_ghcbs(pgd)) {
218 		pr_err("Failed to create 1:1 mapping for the GHCBs!\n");
219 		return 1;
220 	}
221 
222 	/*
223 	 * When making calls to the firmware everything needs to be 1:1
224 	 * mapped and addressable with 32-bit pointers. Map the kernel
225 	 * text and allocate a new stack because we can't rely on the
226 	 * stack pointer being < 4GB.
227 	 */
228 	if (!efi_is_mixed())
229 		return 0;
230 
231 	page = alloc_page(GFP_KERNEL|__GFP_DMA32);
232 	if (!page) {
233 		pr_err("Unable to allocate EFI runtime stack < 4GB\n");
234 		return 1;
235 	}
236 
237 	efi_mixed_mode_stack_pa = page_to_phys(page + 1); /* stack grows down */
238 
239 	npages = (_etext - _text) >> PAGE_SHIFT;
240 	text = __pa(_text);
241 	pfn = text >> PAGE_SHIFT;
242 
243 	pf = _PAGE_ENC;
244 	if (kernel_map_pages_in_pgd(pgd, pfn, text, npages, pf)) {
245 		pr_err("Failed to map kernel text 1:1\n");
246 		return 1;
247 	}
248 
249 	npages = (__end_rodata - __start_rodata) >> PAGE_SHIFT;
250 	rodata = __pa(__start_rodata);
251 	pfn = rodata >> PAGE_SHIFT;
252 
253 	pf = _PAGE_NX | _PAGE_ENC;
254 	if (kernel_map_pages_in_pgd(pgd, pfn, rodata, npages, pf)) {
255 		pr_err("Failed to map kernel rodata 1:1\n");
256 		return 1;
257 	}
258 
259 	return 0;
260 }
261 
262 static void __init __map_region(efi_memory_desc_t *md, u64 va)
263 {
264 	unsigned long flags = _PAGE_RW;
265 	unsigned long pfn;
266 	pgd_t *pgd = efi_mm.pgd;
267 
268 	/*
269 	 * EFI_RUNTIME_SERVICES_CODE regions typically cover PE/COFF
270 	 * executable images in memory that consist of both R-X and
271 	 * RW- sections, so we cannot apply read-only or non-exec
272 	 * permissions just yet. However, modern EFI systems provide
273 	 * a memory attributes table that describes those sections
274 	 * with the appropriate restricted permissions, which are
275 	 * applied in efi_runtime_update_mappings() below. All other
276 	 * regions can be mapped non-executable at this point, with
277 	 * the exception of boot services code regions, but those will
278 	 * be unmapped again entirely in efi_free_boot_services().
279 	 */
280 	if (md->type != EFI_BOOT_SERVICES_CODE &&
281 	    md->type != EFI_RUNTIME_SERVICES_CODE)
282 		flags |= _PAGE_NX;
283 
284 	if (!(md->attribute & EFI_MEMORY_WB))
285 		flags |= _PAGE_PCD;
286 
287 	if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT) &&
288 	    md->type != EFI_MEMORY_MAPPED_IO)
289 		flags |= _PAGE_ENC;
290 
291 	pfn = md->phys_addr >> PAGE_SHIFT;
292 	if (kernel_map_pages_in_pgd(pgd, pfn, va, md->num_pages, flags))
293 		pr_warn("Error mapping PA 0x%llx -> VA 0x%llx!\n",
294 			   md->phys_addr, va);
295 }
296 
297 void __init efi_map_region(efi_memory_desc_t *md)
298 {
299 	unsigned long size = md->num_pages << PAGE_SHIFT;
300 	u64 pa = md->phys_addr;
301 
302 	/*
303 	 * Make sure the 1:1 mappings are present as a catch-all for b0rked
304 	 * firmware which doesn't update all internal pointers after switching
305 	 * to virtual mode and would otherwise crap on us.
306 	 */
307 	__map_region(md, md->phys_addr);
308 
309 	/*
310 	 * Enforce the 1:1 mapping as the default virtual address when
311 	 * booting in EFI mixed mode, because even though we may be
312 	 * running a 64-bit kernel, the firmware may only be 32-bit.
313 	 */
314 	if (efi_is_mixed()) {
315 		md->virt_addr = md->phys_addr;
316 		return;
317 	}
318 
319 	efi_va -= size;
320 
321 	/* Is PA 2M-aligned? */
322 	if (!(pa & (PMD_SIZE - 1))) {
323 		efi_va &= PMD_MASK;
324 	} else {
325 		u64 pa_offset = pa & (PMD_SIZE - 1);
326 		u64 prev_va = efi_va;
327 
328 		/* get us the same offset within this 2M page */
329 		efi_va = (efi_va & PMD_MASK) + pa_offset;
330 
331 		if (efi_va > prev_va)
332 			efi_va -= PMD_SIZE;
333 	}
334 
335 	if (efi_va < EFI_VA_END) {
336 		pr_warn(FW_WARN "VA address range overflow!\n");
337 		return;
338 	}
339 
340 	/* Do the VA map */
341 	__map_region(md, efi_va);
342 	md->virt_addr = efi_va;
343 }
344 
345 /*
346  * kexec kernel will use efi_map_region_fixed to map efi runtime memory ranges.
347  * md->virt_addr is the original virtual address which had been mapped in kexec
348  * 1st kernel.
349  */
350 void __init efi_map_region_fixed(efi_memory_desc_t *md)
351 {
352 	__map_region(md, md->phys_addr);
353 	__map_region(md, md->virt_addr);
354 }
355 
356 void __init parse_efi_setup(u64 phys_addr, u32 data_len)
357 {
358 	efi_setup = phys_addr + sizeof(struct setup_data);
359 }
360 
361 static int __init efi_update_mappings(efi_memory_desc_t *md, unsigned long pf)
362 {
363 	unsigned long pfn;
364 	pgd_t *pgd = efi_mm.pgd;
365 	int err1, err2;
366 
367 	/* Update the 1:1 mapping */
368 	pfn = md->phys_addr >> PAGE_SHIFT;
369 	err1 = kernel_map_pages_in_pgd(pgd, pfn, md->phys_addr, md->num_pages, pf);
370 	if (err1) {
371 		pr_err("Error while updating 1:1 mapping PA 0x%llx -> VA 0x%llx!\n",
372 			   md->phys_addr, md->virt_addr);
373 	}
374 
375 	err2 = kernel_map_pages_in_pgd(pgd, pfn, md->virt_addr, md->num_pages, pf);
376 	if (err2) {
377 		pr_err("Error while updating VA mapping PA 0x%llx -> VA 0x%llx!\n",
378 			   md->phys_addr, md->virt_addr);
379 	}
380 
381 	return err1 || err2;
382 }
383 
384 static int __init efi_update_mem_attr(struct mm_struct *mm, efi_memory_desc_t *md)
385 {
386 	unsigned long pf = 0;
387 
388 	if (md->attribute & EFI_MEMORY_XP)
389 		pf |= _PAGE_NX;
390 
391 	if (!(md->attribute & EFI_MEMORY_RO))
392 		pf |= _PAGE_RW;
393 
394 	if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT))
395 		pf |= _PAGE_ENC;
396 
397 	return efi_update_mappings(md, pf);
398 }
399 
400 void __init efi_runtime_update_mappings(void)
401 {
402 	efi_memory_desc_t *md;
403 
404 	/*
405 	 * Use the EFI Memory Attribute Table for mapping permissions if it
406 	 * exists, since it is intended to supersede EFI_PROPERTIES_TABLE.
407 	 */
408 	if (efi_enabled(EFI_MEM_ATTR)) {
409 		efi_memattr_apply_permissions(NULL, efi_update_mem_attr);
410 		return;
411 	}
412 
413 	/*
414 	 * EFI_MEMORY_ATTRIBUTES_TABLE is intended to replace
415 	 * EFI_PROPERTIES_TABLE. So, use EFI_PROPERTIES_TABLE to update
416 	 * permissions only if EFI_MEMORY_ATTRIBUTES_TABLE is not
417 	 * published by the firmware. Even if we find a buggy implementation of
418 	 * EFI_MEMORY_ATTRIBUTES_TABLE, don't fall back to
419 	 * EFI_PROPERTIES_TABLE, because of the same reason.
420 	 */
421 
422 	if (!efi_enabled(EFI_NX_PE_DATA))
423 		return;
424 
425 	for_each_efi_memory_desc(md) {
426 		unsigned long pf = 0;
427 
428 		if (!(md->attribute & EFI_MEMORY_RUNTIME))
429 			continue;
430 
431 		if (!(md->attribute & EFI_MEMORY_WB))
432 			pf |= _PAGE_PCD;
433 
434 		if ((md->attribute & EFI_MEMORY_XP) ||
435 			(md->type == EFI_RUNTIME_SERVICES_DATA))
436 			pf |= _PAGE_NX;
437 
438 		if (!(md->attribute & EFI_MEMORY_RO) &&
439 			(md->type != EFI_RUNTIME_SERVICES_CODE))
440 			pf |= _PAGE_RW;
441 
442 		if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT))
443 			pf |= _PAGE_ENC;
444 
445 		efi_update_mappings(md, pf);
446 	}
447 }
448 
449 void __init efi_dump_pagetable(void)
450 {
451 #ifdef CONFIG_EFI_PGT_DUMP
452 	ptdump_walk_pgd_level(NULL, &efi_mm);
453 #endif
454 }
455 
456 /*
457  * Makes the calling thread switch to/from efi_mm context. Can be used
458  * in a kernel thread and user context. Preemption needs to remain disabled
459  * while the EFI-mm is borrowed. mmgrab()/mmdrop() is not used because the mm
460  * can not change under us.
461  * It should be ensured that there are no concurrent calls to this function.
462  */
463 void efi_enter_mm(void)
464 {
465 	efi_prev_mm = current->active_mm;
466 	current->active_mm = &efi_mm;
467 	switch_mm(efi_prev_mm, &efi_mm, NULL);
468 }
469 
470 void efi_leave_mm(void)
471 {
472 	current->active_mm = efi_prev_mm;
473 	switch_mm(&efi_mm, efi_prev_mm, NULL);
474 }
475 
476 static DEFINE_SPINLOCK(efi_runtime_lock);
477 
478 /*
479  * DS and ES contain user values.  We need to save them.
480  * The 32-bit EFI code needs a valid DS, ES, and SS.  There's no
481  * need to save the old SS: __KERNEL_DS is always acceptable.
482  */
483 #define __efi_thunk(func, ...)						\
484 ({									\
485 	unsigned short __ds, __es;					\
486 	efi_status_t ____s;						\
487 									\
488 	savesegment(ds, __ds);						\
489 	savesegment(es, __es);						\
490 									\
491 	loadsegment(ss, __KERNEL_DS);					\
492 	loadsegment(ds, __KERNEL_DS);					\
493 	loadsegment(es, __KERNEL_DS);					\
494 									\
495 	____s = efi64_thunk(efi.runtime->mixed_mode.func, __VA_ARGS__);	\
496 									\
497 	loadsegment(ds, __ds);						\
498 	loadsegment(es, __es);						\
499 									\
500 	____s ^= (____s & BIT(31)) | (____s & BIT_ULL(31)) << 32;	\
501 	____s;								\
502 })
503 
504 /*
505  * Switch to the EFI page tables early so that we can access the 1:1
506  * runtime services mappings which are not mapped in any other page
507  * tables.
508  *
509  * Also, disable interrupts because the IDT points to 64-bit handlers,
510  * which aren't going to function correctly when we switch to 32-bit.
511  */
512 #define efi_thunk(func...)						\
513 ({									\
514 	efi_status_t __s;						\
515 									\
516 	arch_efi_call_virt_setup();					\
517 									\
518 	__s = __efi_thunk(func);					\
519 									\
520 	arch_efi_call_virt_teardown();					\
521 									\
522 	__s;								\
523 })
524 
525 static efi_status_t __init __no_sanitize_address
526 efi_thunk_set_virtual_address_map(unsigned long memory_map_size,
527 				  unsigned long descriptor_size,
528 				  u32 descriptor_version,
529 				  efi_memory_desc_t *virtual_map)
530 {
531 	efi_status_t status;
532 	unsigned long flags;
533 
534 	efi_sync_low_kernel_mappings();
535 	local_irq_save(flags);
536 
537 	efi_enter_mm();
538 
539 	status = __efi_thunk(set_virtual_address_map, memory_map_size,
540 			     descriptor_size, descriptor_version, virtual_map);
541 
542 	efi_leave_mm();
543 	local_irq_restore(flags);
544 
545 	return status;
546 }
547 
548 static efi_status_t efi_thunk_get_time(efi_time_t *tm, efi_time_cap_t *tc)
549 {
550 	return EFI_UNSUPPORTED;
551 }
552 
553 static efi_status_t efi_thunk_set_time(efi_time_t *tm)
554 {
555 	return EFI_UNSUPPORTED;
556 }
557 
558 static efi_status_t
559 efi_thunk_get_wakeup_time(efi_bool_t *enabled, efi_bool_t *pending,
560 			  efi_time_t *tm)
561 {
562 	return EFI_UNSUPPORTED;
563 }
564 
565 static efi_status_t
566 efi_thunk_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm)
567 {
568 	return EFI_UNSUPPORTED;
569 }
570 
571 static unsigned long efi_name_size(efi_char16_t *name)
572 {
573 	return ucs2_strsize(name, EFI_VAR_NAME_LEN) + 1;
574 }
575 
576 static efi_status_t
577 efi_thunk_get_variable(efi_char16_t *name, efi_guid_t *vendor,
578 		       u32 *attr, unsigned long *data_size, void *data)
579 {
580 	u8 buf[24] __aligned(8);
581 	efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
582 	efi_status_t status;
583 	u32 phys_name, phys_vendor, phys_attr;
584 	u32 phys_data_size, phys_data;
585 	unsigned long flags;
586 
587 	spin_lock_irqsave(&efi_runtime_lock, flags);
588 
589 	*vnd = *vendor;
590 
591 	phys_data_size = virt_to_phys_or_null(data_size);
592 	phys_vendor = virt_to_phys_or_null(vnd);
593 	phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
594 	phys_attr = virt_to_phys_or_null(attr);
595 	phys_data = virt_to_phys_or_null_size(data, *data_size);
596 
597 	if (!phys_name || (data && !phys_data))
598 		status = EFI_INVALID_PARAMETER;
599 	else
600 		status = efi_thunk(get_variable, phys_name, phys_vendor,
601 				   phys_attr, phys_data_size, phys_data);
602 
603 	spin_unlock_irqrestore(&efi_runtime_lock, flags);
604 
605 	return status;
606 }
607 
608 static efi_status_t
609 efi_thunk_set_variable(efi_char16_t *name, efi_guid_t *vendor,
610 		       u32 attr, unsigned long data_size, void *data)
611 {
612 	u8 buf[24] __aligned(8);
613 	efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
614 	u32 phys_name, phys_vendor, phys_data;
615 	efi_status_t status;
616 	unsigned long flags;
617 
618 	spin_lock_irqsave(&efi_runtime_lock, flags);
619 
620 	*vnd = *vendor;
621 
622 	phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
623 	phys_vendor = virt_to_phys_or_null(vnd);
624 	phys_data = virt_to_phys_or_null_size(data, data_size);
625 
626 	if (!phys_name || (data && !phys_data))
627 		status = EFI_INVALID_PARAMETER;
628 	else
629 		status = efi_thunk(set_variable, phys_name, phys_vendor,
630 				   attr, data_size, phys_data);
631 
632 	spin_unlock_irqrestore(&efi_runtime_lock, flags);
633 
634 	return status;
635 }
636 
637 static efi_status_t
638 efi_thunk_set_variable_nonblocking(efi_char16_t *name, efi_guid_t *vendor,
639 				   u32 attr, unsigned long data_size,
640 				   void *data)
641 {
642 	u8 buf[24] __aligned(8);
643 	efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
644 	u32 phys_name, phys_vendor, phys_data;
645 	efi_status_t status;
646 	unsigned long flags;
647 
648 	if (!spin_trylock_irqsave(&efi_runtime_lock, flags))
649 		return EFI_NOT_READY;
650 
651 	*vnd = *vendor;
652 
653 	phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
654 	phys_vendor = virt_to_phys_or_null(vnd);
655 	phys_data = virt_to_phys_or_null_size(data, data_size);
656 
657 	if (!phys_name || (data && !phys_data))
658 		status = EFI_INVALID_PARAMETER;
659 	else
660 		status = efi_thunk(set_variable, phys_name, phys_vendor,
661 				   attr, data_size, phys_data);
662 
663 	spin_unlock_irqrestore(&efi_runtime_lock, flags);
664 
665 	return status;
666 }
667 
668 static efi_status_t
669 efi_thunk_get_next_variable(unsigned long *name_size,
670 			    efi_char16_t *name,
671 			    efi_guid_t *vendor)
672 {
673 	u8 buf[24] __aligned(8);
674 	efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
675 	efi_status_t status;
676 	u32 phys_name_size, phys_name, phys_vendor;
677 	unsigned long flags;
678 
679 	spin_lock_irqsave(&efi_runtime_lock, flags);
680 
681 	*vnd = *vendor;
682 
683 	phys_name_size = virt_to_phys_or_null(name_size);
684 	phys_vendor = virt_to_phys_or_null(vnd);
685 	phys_name = virt_to_phys_or_null_size(name, *name_size);
686 
687 	if (!phys_name)
688 		status = EFI_INVALID_PARAMETER;
689 	else
690 		status = efi_thunk(get_next_variable, phys_name_size,
691 				   phys_name, phys_vendor);
692 
693 	spin_unlock_irqrestore(&efi_runtime_lock, flags);
694 
695 	*vendor = *vnd;
696 	return status;
697 }
698 
699 static efi_status_t
700 efi_thunk_get_next_high_mono_count(u32 *count)
701 {
702 	return EFI_UNSUPPORTED;
703 }
704 
705 static void
706 efi_thunk_reset_system(int reset_type, efi_status_t status,
707 		       unsigned long data_size, efi_char16_t *data)
708 {
709 	u32 phys_data;
710 	unsigned long flags;
711 
712 	spin_lock_irqsave(&efi_runtime_lock, flags);
713 
714 	phys_data = virt_to_phys_or_null_size(data, data_size);
715 
716 	efi_thunk(reset_system, reset_type, status, data_size, phys_data);
717 
718 	spin_unlock_irqrestore(&efi_runtime_lock, flags);
719 }
720 
721 static efi_status_t
722 efi_thunk_update_capsule(efi_capsule_header_t **capsules,
723 			 unsigned long count, unsigned long sg_list)
724 {
725 	/*
726 	 * To properly support this function we would need to repackage
727 	 * 'capsules' because the firmware doesn't understand 64-bit
728 	 * pointers.
729 	 */
730 	return EFI_UNSUPPORTED;
731 }
732 
733 static efi_status_t
734 efi_thunk_query_variable_info(u32 attr, u64 *storage_space,
735 			      u64 *remaining_space,
736 			      u64 *max_variable_size)
737 {
738 	efi_status_t status;
739 	u32 phys_storage, phys_remaining, phys_max;
740 	unsigned long flags;
741 
742 	if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
743 		return EFI_UNSUPPORTED;
744 
745 	spin_lock_irqsave(&efi_runtime_lock, flags);
746 
747 	phys_storage = virt_to_phys_or_null(storage_space);
748 	phys_remaining = virt_to_phys_or_null(remaining_space);
749 	phys_max = virt_to_phys_or_null(max_variable_size);
750 
751 	status = efi_thunk(query_variable_info, attr, phys_storage,
752 			   phys_remaining, phys_max);
753 
754 	spin_unlock_irqrestore(&efi_runtime_lock, flags);
755 
756 	return status;
757 }
758 
759 static efi_status_t
760 efi_thunk_query_variable_info_nonblocking(u32 attr, u64 *storage_space,
761 					  u64 *remaining_space,
762 					  u64 *max_variable_size)
763 {
764 	efi_status_t status;
765 	u32 phys_storage, phys_remaining, phys_max;
766 	unsigned long flags;
767 
768 	if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
769 		return EFI_UNSUPPORTED;
770 
771 	if (!spin_trylock_irqsave(&efi_runtime_lock, flags))
772 		return EFI_NOT_READY;
773 
774 	phys_storage = virt_to_phys_or_null(storage_space);
775 	phys_remaining = virt_to_phys_or_null(remaining_space);
776 	phys_max = virt_to_phys_or_null(max_variable_size);
777 
778 	status = efi_thunk(query_variable_info, attr, phys_storage,
779 			   phys_remaining, phys_max);
780 
781 	spin_unlock_irqrestore(&efi_runtime_lock, flags);
782 
783 	return status;
784 }
785 
786 static efi_status_t
787 efi_thunk_query_capsule_caps(efi_capsule_header_t **capsules,
788 			     unsigned long count, u64 *max_size,
789 			     int *reset_type)
790 {
791 	/*
792 	 * To properly support this function we would need to repackage
793 	 * 'capsules' because the firmware doesn't understand 64-bit
794 	 * pointers.
795 	 */
796 	return EFI_UNSUPPORTED;
797 }
798 
799 void __init efi_thunk_runtime_setup(void)
800 {
801 	if (!IS_ENABLED(CONFIG_EFI_MIXED))
802 		return;
803 
804 	efi.get_time = efi_thunk_get_time;
805 	efi.set_time = efi_thunk_set_time;
806 	efi.get_wakeup_time = efi_thunk_get_wakeup_time;
807 	efi.set_wakeup_time = efi_thunk_set_wakeup_time;
808 	efi.get_variable = efi_thunk_get_variable;
809 	efi.get_next_variable = efi_thunk_get_next_variable;
810 	efi.set_variable = efi_thunk_set_variable;
811 	efi.set_variable_nonblocking = efi_thunk_set_variable_nonblocking;
812 	efi.get_next_high_mono_count = efi_thunk_get_next_high_mono_count;
813 	efi.reset_system = efi_thunk_reset_system;
814 	efi.query_variable_info = efi_thunk_query_variable_info;
815 	efi.query_variable_info_nonblocking = efi_thunk_query_variable_info_nonblocking;
816 	efi.update_capsule = efi_thunk_update_capsule;
817 	efi.query_capsule_caps = efi_thunk_query_capsule_caps;
818 }
819 
820 efi_status_t __init __no_sanitize_address
821 efi_set_virtual_address_map(unsigned long memory_map_size,
822 			    unsigned long descriptor_size,
823 			    u32 descriptor_version,
824 			    efi_memory_desc_t *virtual_map,
825 			    unsigned long systab_phys)
826 {
827 	const efi_system_table_t *systab = (efi_system_table_t *)systab_phys;
828 	efi_status_t status;
829 	unsigned long flags;
830 
831 	if (efi_is_mixed())
832 		return efi_thunk_set_virtual_address_map(memory_map_size,
833 							 descriptor_size,
834 							 descriptor_version,
835 							 virtual_map);
836 	efi_enter_mm();
837 
838 	efi_fpu_begin();
839 
840 	/* Disable interrupts around EFI calls: */
841 	local_irq_save(flags);
842 	status = efi_call(efi.runtime->set_virtual_address_map,
843 			  memory_map_size, descriptor_size,
844 			  descriptor_version, virtual_map);
845 	local_irq_restore(flags);
846 
847 	efi_fpu_end();
848 
849 	/* grab the virtually remapped EFI runtime services table pointer */
850 	efi.runtime = READ_ONCE(systab->runtime);
851 
852 	efi_leave_mm();
853 
854 	return status;
855 }
856