xref: /openbmc/linux/arch/x86/platform/efi/efi_64.c (revision 15e3ae36)
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/mem_encrypt.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/pgtable.h>
43 #include <asm/tlbflush.h>
44 #include <asm/proto.h>
45 #include <asm/efi.h>
46 #include <asm/cacheflush.h>
47 #include <asm/fixmap.h>
48 #include <asm/realmode.h>
49 #include <asm/time.h>
50 #include <asm/pgalloc.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 
58 struct efi_scratch efi_scratch;
59 
60 EXPORT_SYMBOL_GPL(efi_mm);
61 
62 /*
63  * We need our own copy of the higher levels of the page tables
64  * because we want to avoid inserting EFI region mappings (EFI_VA_END
65  * to EFI_VA_START) into the standard kernel page tables. Everything
66  * else can be shared, see efi_sync_low_kernel_mappings().
67  *
68  * We don't want the pgd on the pgd_list and cannot use pgd_alloc() for the
69  * allocation.
70  */
71 int __init efi_alloc_page_tables(void)
72 {
73 	pgd_t *pgd, *efi_pgd;
74 	p4d_t *p4d;
75 	pud_t *pud;
76 	gfp_t gfp_mask;
77 
78 	if (efi_have_uv1_memmap())
79 		return 0;
80 
81 	gfp_mask = GFP_KERNEL | __GFP_ZERO;
82 	efi_pgd = (pgd_t *)__get_free_pages(gfp_mask, PGD_ALLOCATION_ORDER);
83 	if (!efi_pgd)
84 		return -ENOMEM;
85 
86 	pgd = efi_pgd + pgd_index(EFI_VA_END);
87 	p4d = p4d_alloc(&init_mm, pgd, EFI_VA_END);
88 	if (!p4d) {
89 		free_page((unsigned long)efi_pgd);
90 		return -ENOMEM;
91 	}
92 
93 	pud = pud_alloc(&init_mm, p4d, EFI_VA_END);
94 	if (!pud) {
95 		if (pgtable_l5_enabled())
96 			free_page((unsigned long) pgd_page_vaddr(*pgd));
97 		free_pages((unsigned long)efi_pgd, PGD_ALLOCATION_ORDER);
98 		return -ENOMEM;
99 	}
100 
101 	efi_mm.pgd = efi_pgd;
102 	mm_init_cpumask(&efi_mm);
103 	init_new_context(NULL, &efi_mm);
104 
105 	return 0;
106 }
107 
108 /*
109  * Add low kernel mappings for passing arguments to EFI functions.
110  */
111 void efi_sync_low_kernel_mappings(void)
112 {
113 	unsigned num_entries;
114 	pgd_t *pgd_k, *pgd_efi;
115 	p4d_t *p4d_k, *p4d_efi;
116 	pud_t *pud_k, *pud_efi;
117 	pgd_t *efi_pgd = efi_mm.pgd;
118 
119 	if (efi_have_uv1_memmap())
120 		return;
121 
122 	/*
123 	 * We can share all PGD entries apart from the one entry that
124 	 * covers the EFI runtime mapping space.
125 	 *
126 	 * Make sure the EFI runtime region mappings are guaranteed to
127 	 * only span a single PGD entry and that the entry also maps
128 	 * other important kernel regions.
129 	 */
130 	MAYBE_BUILD_BUG_ON(pgd_index(EFI_VA_END) != pgd_index(MODULES_END));
131 	MAYBE_BUILD_BUG_ON((EFI_VA_START & PGDIR_MASK) !=
132 			(EFI_VA_END & PGDIR_MASK));
133 
134 	pgd_efi = efi_pgd + pgd_index(PAGE_OFFSET);
135 	pgd_k = pgd_offset_k(PAGE_OFFSET);
136 
137 	num_entries = pgd_index(EFI_VA_END) - pgd_index(PAGE_OFFSET);
138 	memcpy(pgd_efi, pgd_k, sizeof(pgd_t) * num_entries);
139 
140 	/*
141 	 * As with PGDs, we share all P4D entries apart from the one entry
142 	 * that covers the EFI runtime mapping space.
143 	 */
144 	BUILD_BUG_ON(p4d_index(EFI_VA_END) != p4d_index(MODULES_END));
145 	BUILD_BUG_ON((EFI_VA_START & P4D_MASK) != (EFI_VA_END & P4D_MASK));
146 
147 	pgd_efi = efi_pgd + pgd_index(EFI_VA_END);
148 	pgd_k = pgd_offset_k(EFI_VA_END);
149 	p4d_efi = p4d_offset(pgd_efi, 0);
150 	p4d_k = p4d_offset(pgd_k, 0);
151 
152 	num_entries = p4d_index(EFI_VA_END);
153 	memcpy(p4d_efi, p4d_k, sizeof(p4d_t) * num_entries);
154 
155 	/*
156 	 * We share all the PUD entries apart from those that map the
157 	 * EFI regions. Copy around them.
158 	 */
159 	BUILD_BUG_ON((EFI_VA_START & ~PUD_MASK) != 0);
160 	BUILD_BUG_ON((EFI_VA_END & ~PUD_MASK) != 0);
161 
162 	p4d_efi = p4d_offset(pgd_efi, EFI_VA_END);
163 	p4d_k = p4d_offset(pgd_k, EFI_VA_END);
164 	pud_efi = pud_offset(p4d_efi, 0);
165 	pud_k = pud_offset(p4d_k, 0);
166 
167 	num_entries = pud_index(EFI_VA_END);
168 	memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
169 
170 	pud_efi = pud_offset(p4d_efi, EFI_VA_START);
171 	pud_k = pud_offset(p4d_k, EFI_VA_START);
172 
173 	num_entries = PTRS_PER_PUD - pud_index(EFI_VA_START);
174 	memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
175 }
176 
177 /*
178  * Wrapper for slow_virt_to_phys() that handles NULL addresses.
179  */
180 static inline phys_addr_t
181 virt_to_phys_or_null_size(void *va, unsigned long size)
182 {
183 	phys_addr_t pa;
184 
185 	if (!va)
186 		return 0;
187 
188 	if (virt_addr_valid(va))
189 		return virt_to_phys(va);
190 
191 	pa = slow_virt_to_phys(va);
192 
193 	/* check if the object crosses a page boundary */
194 	if (WARN_ON((pa ^ (pa + size - 1)) & PAGE_MASK))
195 		return 0;
196 
197 	return pa;
198 }
199 
200 #define virt_to_phys_or_null(addr)				\
201 	virt_to_phys_or_null_size((addr), sizeof(*(addr)))
202 
203 int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)
204 {
205 	unsigned long pfn, text, pf;
206 	struct page *page;
207 	unsigned npages;
208 	pgd_t *pgd = efi_mm.pgd;
209 
210 	if (efi_have_uv1_memmap())
211 		return 0;
212 
213 	/*
214 	 * It can happen that the physical address of new_memmap lands in memory
215 	 * which is not mapped in the EFI page table. Therefore we need to go
216 	 * and ident-map those pages containing the map before calling
217 	 * phys_efi_set_virtual_address_map().
218 	 */
219 	pfn = pa_memmap >> PAGE_SHIFT;
220 	pf = _PAGE_NX | _PAGE_RW | _PAGE_ENC;
221 	if (kernel_map_pages_in_pgd(pgd, pfn, pa_memmap, num_pages, pf)) {
222 		pr_err("Error ident-mapping new memmap (0x%lx)!\n", pa_memmap);
223 		return 1;
224 	}
225 
226 	/*
227 	 * Certain firmware versions are way too sentimential and still believe
228 	 * they are exclusive and unquestionable owners of the first physical page,
229 	 * even though they explicitly mark it as EFI_CONVENTIONAL_MEMORY
230 	 * (but then write-access it later during SetVirtualAddressMap()).
231 	 *
232 	 * Create a 1:1 mapping for this page, to avoid triple faults during early
233 	 * boot with such firmware. We are free to hand this page to the BIOS,
234 	 * as trim_bios_range() will reserve the first page and isolate it away
235 	 * from memory allocators anyway.
236 	 */
237 	if (kernel_map_pages_in_pgd(pgd, 0x0, 0x0, 1, pf)) {
238 		pr_err("Failed to create 1:1 mapping for the first page!\n");
239 		return 1;
240 	}
241 
242 	/*
243 	 * When making calls to the firmware everything needs to be 1:1
244 	 * mapped and addressable with 32-bit pointers. Map the kernel
245 	 * text and allocate a new stack because we can't rely on the
246 	 * stack pointer being < 4GB.
247 	 */
248 	if (!efi_is_mixed())
249 		return 0;
250 
251 	page = alloc_page(GFP_KERNEL|__GFP_DMA32);
252 	if (!page) {
253 		pr_err("Unable to allocate EFI runtime stack < 4GB\n");
254 		return 1;
255 	}
256 
257 	efi_scratch.phys_stack = page_to_phys(page + 1); /* stack grows down */
258 
259 	npages = (__end_rodata_aligned - _text) >> PAGE_SHIFT;
260 	text = __pa(_text);
261 	pfn = text >> PAGE_SHIFT;
262 
263 	pf = _PAGE_ENC;
264 	if (kernel_map_pages_in_pgd(pgd, pfn, text, npages, pf)) {
265 		pr_err("Failed to map kernel text 1:1\n");
266 		return 1;
267 	}
268 
269 	return 0;
270 }
271 
272 static void __init __map_region(efi_memory_desc_t *md, u64 va)
273 {
274 	unsigned long flags = _PAGE_RW;
275 	unsigned long pfn;
276 	pgd_t *pgd = efi_mm.pgd;
277 
278 	/*
279 	 * EFI_RUNTIME_SERVICES_CODE regions typically cover PE/COFF
280 	 * executable images in memory that consist of both R-X and
281 	 * RW- sections, so we cannot apply read-only or non-exec
282 	 * permissions just yet. However, modern EFI systems provide
283 	 * a memory attributes table that describes those sections
284 	 * with the appropriate restricted permissions, which are
285 	 * applied in efi_runtime_update_mappings() below. All other
286 	 * regions can be mapped non-executable at this point, with
287 	 * the exception of boot services code regions, but those will
288 	 * be unmapped again entirely in efi_free_boot_services().
289 	 */
290 	if (md->type != EFI_BOOT_SERVICES_CODE &&
291 	    md->type != EFI_RUNTIME_SERVICES_CODE)
292 		flags |= _PAGE_NX;
293 
294 	if (!(md->attribute & EFI_MEMORY_WB))
295 		flags |= _PAGE_PCD;
296 
297 	if (sev_active() && md->type != EFI_MEMORY_MAPPED_IO)
298 		flags |= _PAGE_ENC;
299 
300 	pfn = md->phys_addr >> PAGE_SHIFT;
301 	if (kernel_map_pages_in_pgd(pgd, pfn, va, md->num_pages, flags))
302 		pr_warn("Error mapping PA 0x%llx -> VA 0x%llx!\n",
303 			   md->phys_addr, va);
304 }
305 
306 void __init efi_map_region(efi_memory_desc_t *md)
307 {
308 	unsigned long size = md->num_pages << PAGE_SHIFT;
309 	u64 pa = md->phys_addr;
310 
311 	if (efi_have_uv1_memmap())
312 		return old_map_region(md);
313 
314 	/*
315 	 * Make sure the 1:1 mappings are present as a catch-all for b0rked
316 	 * firmware which doesn't update all internal pointers after switching
317 	 * to virtual mode and would otherwise crap on us.
318 	 */
319 	__map_region(md, md->phys_addr);
320 
321 	/*
322 	 * Enforce the 1:1 mapping as the default virtual address when
323 	 * booting in EFI mixed mode, because even though we may be
324 	 * running a 64-bit kernel, the firmware may only be 32-bit.
325 	 */
326 	if (efi_is_mixed()) {
327 		md->virt_addr = md->phys_addr;
328 		return;
329 	}
330 
331 	efi_va -= size;
332 
333 	/* Is PA 2M-aligned? */
334 	if (!(pa & (PMD_SIZE - 1))) {
335 		efi_va &= PMD_MASK;
336 	} else {
337 		u64 pa_offset = pa & (PMD_SIZE - 1);
338 		u64 prev_va = efi_va;
339 
340 		/* get us the same offset within this 2M page */
341 		efi_va = (efi_va & PMD_MASK) + pa_offset;
342 
343 		if (efi_va > prev_va)
344 			efi_va -= PMD_SIZE;
345 	}
346 
347 	if (efi_va < EFI_VA_END) {
348 		pr_warn(FW_WARN "VA address range overflow!\n");
349 		return;
350 	}
351 
352 	/* Do the VA map */
353 	__map_region(md, efi_va);
354 	md->virt_addr = efi_va;
355 }
356 
357 /*
358  * kexec kernel will use efi_map_region_fixed to map efi runtime memory ranges.
359  * md->virt_addr is the original virtual address which had been mapped in kexec
360  * 1st kernel.
361  */
362 void __init efi_map_region_fixed(efi_memory_desc_t *md)
363 {
364 	__map_region(md, md->phys_addr);
365 	__map_region(md, md->virt_addr);
366 }
367 
368 void __init parse_efi_setup(u64 phys_addr, u32 data_len)
369 {
370 	efi_setup = phys_addr + sizeof(struct setup_data);
371 }
372 
373 static int __init efi_update_mappings(efi_memory_desc_t *md, unsigned long pf)
374 {
375 	unsigned long pfn;
376 	pgd_t *pgd = efi_mm.pgd;
377 	int err1, err2;
378 
379 	/* Update the 1:1 mapping */
380 	pfn = md->phys_addr >> PAGE_SHIFT;
381 	err1 = kernel_map_pages_in_pgd(pgd, pfn, md->phys_addr, md->num_pages, pf);
382 	if (err1) {
383 		pr_err("Error while updating 1:1 mapping PA 0x%llx -> VA 0x%llx!\n",
384 			   md->phys_addr, md->virt_addr);
385 	}
386 
387 	err2 = kernel_map_pages_in_pgd(pgd, pfn, md->virt_addr, md->num_pages, pf);
388 	if (err2) {
389 		pr_err("Error while updating VA mapping PA 0x%llx -> VA 0x%llx!\n",
390 			   md->phys_addr, md->virt_addr);
391 	}
392 
393 	return err1 || err2;
394 }
395 
396 static int __init efi_update_mem_attr(struct mm_struct *mm, efi_memory_desc_t *md)
397 {
398 	unsigned long pf = 0;
399 
400 	if (md->attribute & EFI_MEMORY_XP)
401 		pf |= _PAGE_NX;
402 
403 	if (!(md->attribute & EFI_MEMORY_RO))
404 		pf |= _PAGE_RW;
405 
406 	if (sev_active())
407 		pf |= _PAGE_ENC;
408 
409 	return efi_update_mappings(md, pf);
410 }
411 
412 void __init efi_runtime_update_mappings(void)
413 {
414 	efi_memory_desc_t *md;
415 
416 	if (efi_have_uv1_memmap()) {
417 		if (__supported_pte_mask & _PAGE_NX)
418 			runtime_code_page_mkexec();
419 		return;
420 	}
421 
422 	/*
423 	 * Use the EFI Memory Attribute Table for mapping permissions if it
424 	 * exists, since it is intended to supersede EFI_PROPERTIES_TABLE.
425 	 */
426 	if (efi_enabled(EFI_MEM_ATTR)) {
427 		efi_memattr_apply_permissions(NULL, efi_update_mem_attr);
428 		return;
429 	}
430 
431 	/*
432 	 * EFI_MEMORY_ATTRIBUTES_TABLE is intended to replace
433 	 * EFI_PROPERTIES_TABLE. So, use EFI_PROPERTIES_TABLE to update
434 	 * permissions only if EFI_MEMORY_ATTRIBUTES_TABLE is not
435 	 * published by the firmware. Even if we find a buggy implementation of
436 	 * EFI_MEMORY_ATTRIBUTES_TABLE, don't fall back to
437 	 * EFI_PROPERTIES_TABLE, because of the same reason.
438 	 */
439 
440 	if (!efi_enabled(EFI_NX_PE_DATA))
441 		return;
442 
443 	for_each_efi_memory_desc(md) {
444 		unsigned long pf = 0;
445 
446 		if (!(md->attribute & EFI_MEMORY_RUNTIME))
447 			continue;
448 
449 		if (!(md->attribute & EFI_MEMORY_WB))
450 			pf |= _PAGE_PCD;
451 
452 		if ((md->attribute & EFI_MEMORY_XP) ||
453 			(md->type == EFI_RUNTIME_SERVICES_DATA))
454 			pf |= _PAGE_NX;
455 
456 		if (!(md->attribute & EFI_MEMORY_RO) &&
457 			(md->type != EFI_RUNTIME_SERVICES_CODE))
458 			pf |= _PAGE_RW;
459 
460 		if (sev_active())
461 			pf |= _PAGE_ENC;
462 
463 		efi_update_mappings(md, pf);
464 	}
465 }
466 
467 void __init efi_dump_pagetable(void)
468 {
469 #ifdef CONFIG_EFI_PGT_DUMP
470 	if (efi_have_uv1_memmap())
471 		ptdump_walk_pgd_level(NULL, &init_mm);
472 	else
473 		ptdump_walk_pgd_level(NULL, &efi_mm);
474 #endif
475 }
476 
477 /*
478  * Makes the calling thread switch to/from efi_mm context. Can be used
479  * in a kernel thread and user context. Preemption needs to remain disabled
480  * while the EFI-mm is borrowed. mmgrab()/mmdrop() is not used because the mm
481  * can not change under us.
482  * It should be ensured that there are no concurent calls to this function.
483  */
484 void efi_switch_mm(struct mm_struct *mm)
485 {
486 	efi_scratch.prev_mm = current->active_mm;
487 	current->active_mm = mm;
488 	switch_mm(efi_scratch.prev_mm, mm, NULL);
489 }
490 
491 static DEFINE_SPINLOCK(efi_runtime_lock);
492 
493 /*
494  * DS and ES contain user values.  We need to save them.
495  * The 32-bit EFI code needs a valid DS, ES, and SS.  There's no
496  * need to save the old SS: __KERNEL_DS is always acceptable.
497  */
498 #define __efi_thunk(func, ...)						\
499 ({									\
500 	unsigned short __ds, __es;					\
501 	efi_status_t ____s;						\
502 									\
503 	savesegment(ds, __ds);						\
504 	savesegment(es, __es);						\
505 									\
506 	loadsegment(ss, __KERNEL_DS);					\
507 	loadsegment(ds, __KERNEL_DS);					\
508 	loadsegment(es, __KERNEL_DS);					\
509 									\
510 	____s = efi64_thunk(efi.runtime->mixed_mode.func, __VA_ARGS__);	\
511 									\
512 	loadsegment(ds, __ds);						\
513 	loadsegment(es, __es);						\
514 									\
515 	____s ^= (____s & BIT(31)) | (____s & BIT_ULL(31)) << 32;	\
516 	____s;								\
517 })
518 
519 /*
520  * Switch to the EFI page tables early so that we can access the 1:1
521  * runtime services mappings which are not mapped in any other page
522  * tables.
523  *
524  * Also, disable interrupts because the IDT points to 64-bit handlers,
525  * which aren't going to function correctly when we switch to 32-bit.
526  */
527 #define efi_thunk(func...)						\
528 ({									\
529 	efi_status_t __s;						\
530 									\
531 	arch_efi_call_virt_setup();					\
532 									\
533 	__s = __efi_thunk(func);					\
534 									\
535 	arch_efi_call_virt_teardown();					\
536 									\
537 	__s;								\
538 })
539 
540 static efi_status_t __init __no_sanitize_address
541 efi_thunk_set_virtual_address_map(unsigned long memory_map_size,
542 				  unsigned long descriptor_size,
543 				  u32 descriptor_version,
544 				  efi_memory_desc_t *virtual_map)
545 {
546 	efi_status_t status;
547 	unsigned long flags;
548 
549 	efi_sync_low_kernel_mappings();
550 	local_irq_save(flags);
551 
552 	efi_switch_mm(&efi_mm);
553 
554 	status = __efi_thunk(set_virtual_address_map, memory_map_size,
555 			     descriptor_size, descriptor_version, virtual_map);
556 
557 	efi_switch_mm(efi_scratch.prev_mm);
558 	local_irq_restore(flags);
559 
560 	return status;
561 }
562 
563 static efi_status_t efi_thunk_get_time(efi_time_t *tm, efi_time_cap_t *tc)
564 {
565 	return EFI_UNSUPPORTED;
566 }
567 
568 static efi_status_t efi_thunk_set_time(efi_time_t *tm)
569 {
570 	return EFI_UNSUPPORTED;
571 }
572 
573 static efi_status_t
574 efi_thunk_get_wakeup_time(efi_bool_t *enabled, efi_bool_t *pending,
575 			  efi_time_t *tm)
576 {
577 	return EFI_UNSUPPORTED;
578 }
579 
580 static efi_status_t
581 efi_thunk_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm)
582 {
583 	return EFI_UNSUPPORTED;
584 }
585 
586 static unsigned long efi_name_size(efi_char16_t *name)
587 {
588 	return ucs2_strsize(name, EFI_VAR_NAME_LEN) + 1;
589 }
590 
591 static efi_status_t
592 efi_thunk_get_variable(efi_char16_t *name, efi_guid_t *vendor,
593 		       u32 *attr, unsigned long *data_size, void *data)
594 {
595 	u8 buf[24] __aligned(8);
596 	efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
597 	efi_status_t status;
598 	u32 phys_name, phys_vendor, phys_attr;
599 	u32 phys_data_size, phys_data;
600 	unsigned long flags;
601 
602 	spin_lock_irqsave(&efi_runtime_lock, flags);
603 
604 	*vnd = *vendor;
605 
606 	phys_data_size = virt_to_phys_or_null(data_size);
607 	phys_vendor = virt_to_phys_or_null(vnd);
608 	phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
609 	phys_attr = virt_to_phys_or_null(attr);
610 	phys_data = virt_to_phys_or_null_size(data, *data_size);
611 
612 	if (!phys_name || (data && !phys_data))
613 		status = EFI_INVALID_PARAMETER;
614 	else
615 		status = efi_thunk(get_variable, phys_name, phys_vendor,
616 				   phys_attr, phys_data_size, phys_data);
617 
618 	spin_unlock_irqrestore(&efi_runtime_lock, flags);
619 
620 	return status;
621 }
622 
623 static efi_status_t
624 efi_thunk_set_variable(efi_char16_t *name, efi_guid_t *vendor,
625 		       u32 attr, unsigned long data_size, void *data)
626 {
627 	u8 buf[24] __aligned(8);
628 	efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
629 	u32 phys_name, phys_vendor, phys_data;
630 	efi_status_t status;
631 	unsigned long flags;
632 
633 	spin_lock_irqsave(&efi_runtime_lock, flags);
634 
635 	*vnd = *vendor;
636 
637 	phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
638 	phys_vendor = virt_to_phys_or_null(vnd);
639 	phys_data = virt_to_phys_or_null_size(data, data_size);
640 
641 	if (!phys_name || !phys_data)
642 		status = EFI_INVALID_PARAMETER;
643 	else
644 		status = efi_thunk(set_variable, phys_name, phys_vendor,
645 				   attr, data_size, phys_data);
646 
647 	spin_unlock_irqrestore(&efi_runtime_lock, flags);
648 
649 	return status;
650 }
651 
652 static efi_status_t
653 efi_thunk_set_variable_nonblocking(efi_char16_t *name, efi_guid_t *vendor,
654 				   u32 attr, unsigned long data_size,
655 				   void *data)
656 {
657 	u8 buf[24] __aligned(8);
658 	efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
659 	u32 phys_name, phys_vendor, phys_data;
660 	efi_status_t status;
661 	unsigned long flags;
662 
663 	if (!spin_trylock_irqsave(&efi_runtime_lock, flags))
664 		return EFI_NOT_READY;
665 
666 	*vnd = *vendor;
667 
668 	phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
669 	phys_vendor = virt_to_phys_or_null(vnd);
670 	phys_data = virt_to_phys_or_null_size(data, data_size);
671 
672 	if (!phys_name || !phys_data)
673 		status = EFI_INVALID_PARAMETER;
674 	else
675 		status = efi_thunk(set_variable, phys_name, phys_vendor,
676 				   attr, data_size, phys_data);
677 
678 	spin_unlock_irqrestore(&efi_runtime_lock, flags);
679 
680 	return status;
681 }
682 
683 static efi_status_t
684 efi_thunk_get_next_variable(unsigned long *name_size,
685 			    efi_char16_t *name,
686 			    efi_guid_t *vendor)
687 {
688 	u8 buf[24] __aligned(8);
689 	efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
690 	efi_status_t status;
691 	u32 phys_name_size, phys_name, phys_vendor;
692 	unsigned long flags;
693 
694 	spin_lock_irqsave(&efi_runtime_lock, flags);
695 
696 	*vnd = *vendor;
697 
698 	phys_name_size = virt_to_phys_or_null(name_size);
699 	phys_vendor = virt_to_phys_or_null(vnd);
700 	phys_name = virt_to_phys_or_null_size(name, *name_size);
701 
702 	if (!phys_name)
703 		status = EFI_INVALID_PARAMETER;
704 	else
705 		status = efi_thunk(get_next_variable, phys_name_size,
706 				   phys_name, phys_vendor);
707 
708 	spin_unlock_irqrestore(&efi_runtime_lock, flags);
709 
710 	*vendor = *vnd;
711 	return status;
712 }
713 
714 static efi_status_t
715 efi_thunk_get_next_high_mono_count(u32 *count)
716 {
717 	return EFI_UNSUPPORTED;
718 }
719 
720 static void
721 efi_thunk_reset_system(int reset_type, efi_status_t status,
722 		       unsigned long data_size, efi_char16_t *data)
723 {
724 	u32 phys_data;
725 	unsigned long flags;
726 
727 	spin_lock_irqsave(&efi_runtime_lock, flags);
728 
729 	phys_data = virt_to_phys_or_null_size(data, data_size);
730 
731 	efi_thunk(reset_system, reset_type, status, data_size, phys_data);
732 
733 	spin_unlock_irqrestore(&efi_runtime_lock, flags);
734 }
735 
736 static efi_status_t
737 efi_thunk_update_capsule(efi_capsule_header_t **capsules,
738 			 unsigned long count, unsigned long sg_list)
739 {
740 	/*
741 	 * To properly support this function we would need to repackage
742 	 * 'capsules' because the firmware doesn't understand 64-bit
743 	 * pointers.
744 	 */
745 	return EFI_UNSUPPORTED;
746 }
747 
748 static efi_status_t
749 efi_thunk_query_variable_info(u32 attr, u64 *storage_space,
750 			      u64 *remaining_space,
751 			      u64 *max_variable_size)
752 {
753 	efi_status_t status;
754 	u32 phys_storage, phys_remaining, phys_max;
755 	unsigned long flags;
756 
757 	if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
758 		return EFI_UNSUPPORTED;
759 
760 	spin_lock_irqsave(&efi_runtime_lock, flags);
761 
762 	phys_storage = virt_to_phys_or_null(storage_space);
763 	phys_remaining = virt_to_phys_or_null(remaining_space);
764 	phys_max = virt_to_phys_or_null(max_variable_size);
765 
766 	status = efi_thunk(query_variable_info, attr, phys_storage,
767 			   phys_remaining, phys_max);
768 
769 	spin_unlock_irqrestore(&efi_runtime_lock, flags);
770 
771 	return status;
772 }
773 
774 static efi_status_t
775 efi_thunk_query_variable_info_nonblocking(u32 attr, u64 *storage_space,
776 					  u64 *remaining_space,
777 					  u64 *max_variable_size)
778 {
779 	efi_status_t status;
780 	u32 phys_storage, phys_remaining, phys_max;
781 	unsigned long flags;
782 
783 	if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
784 		return EFI_UNSUPPORTED;
785 
786 	if (!spin_trylock_irqsave(&efi_runtime_lock, flags))
787 		return EFI_NOT_READY;
788 
789 	phys_storage = virt_to_phys_or_null(storage_space);
790 	phys_remaining = virt_to_phys_or_null(remaining_space);
791 	phys_max = virt_to_phys_or_null(max_variable_size);
792 
793 	status = efi_thunk(query_variable_info, attr, phys_storage,
794 			   phys_remaining, phys_max);
795 
796 	spin_unlock_irqrestore(&efi_runtime_lock, flags);
797 
798 	return status;
799 }
800 
801 static efi_status_t
802 efi_thunk_query_capsule_caps(efi_capsule_header_t **capsules,
803 			     unsigned long count, u64 *max_size,
804 			     int *reset_type)
805 {
806 	/*
807 	 * To properly support this function we would need to repackage
808 	 * 'capsules' because the firmware doesn't understand 64-bit
809 	 * pointers.
810 	 */
811 	return EFI_UNSUPPORTED;
812 }
813 
814 void __init efi_thunk_runtime_setup(void)
815 {
816 	if (!IS_ENABLED(CONFIG_EFI_MIXED))
817 		return;
818 
819 	efi.get_time = efi_thunk_get_time;
820 	efi.set_time = efi_thunk_set_time;
821 	efi.get_wakeup_time = efi_thunk_get_wakeup_time;
822 	efi.set_wakeup_time = efi_thunk_set_wakeup_time;
823 	efi.get_variable = efi_thunk_get_variable;
824 	efi.get_next_variable = efi_thunk_get_next_variable;
825 	efi.set_variable = efi_thunk_set_variable;
826 	efi.set_variable_nonblocking = efi_thunk_set_variable_nonblocking;
827 	efi.get_next_high_mono_count = efi_thunk_get_next_high_mono_count;
828 	efi.reset_system = efi_thunk_reset_system;
829 	efi.query_variable_info = efi_thunk_query_variable_info;
830 	efi.query_variable_info_nonblocking = efi_thunk_query_variable_info_nonblocking;
831 	efi.update_capsule = efi_thunk_update_capsule;
832 	efi.query_capsule_caps = efi_thunk_query_capsule_caps;
833 }
834 
835 efi_status_t __init __no_sanitize_address
836 efi_set_virtual_address_map(unsigned long memory_map_size,
837 			    unsigned long descriptor_size,
838 			    u32 descriptor_version,
839 			    efi_memory_desc_t *virtual_map,
840 			    unsigned long systab_phys)
841 {
842 	const efi_system_table_t *systab = (efi_system_table_t *)systab_phys;
843 	efi_status_t status;
844 	unsigned long flags;
845 	pgd_t *save_pgd = NULL;
846 
847 	if (efi_is_mixed())
848 		return efi_thunk_set_virtual_address_map(memory_map_size,
849 							 descriptor_size,
850 							 descriptor_version,
851 							 virtual_map);
852 
853 	if (efi_have_uv1_memmap()) {
854 		save_pgd = efi_uv1_memmap_phys_prolog();
855 		if (!save_pgd)
856 			return EFI_ABORTED;
857 	} else {
858 		efi_switch_mm(&efi_mm);
859 	}
860 
861 	kernel_fpu_begin();
862 
863 	/* Disable interrupts around EFI calls: */
864 	local_irq_save(flags);
865 	status = efi_call(efi.runtime->set_virtual_address_map,
866 			  memory_map_size, descriptor_size,
867 			  descriptor_version, virtual_map);
868 	local_irq_restore(flags);
869 
870 	kernel_fpu_end();
871 
872 	/* grab the virtually remapped EFI runtime services table pointer */
873 	efi.runtime = READ_ONCE(systab->runtime);
874 
875 	if (save_pgd)
876 		efi_uv1_memmap_phys_epilog(save_pgd);
877 	else
878 		efi_switch_mm(efi_scratch.prev_mm);
879 
880 	return status;
881 }
882