xref: /openbmc/linux/arch/x86/platform/efi/efi_64.c (revision b58c6630)
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, rodata;
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 = (_etext - _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 	npages = (__end_rodata - __start_rodata) >> PAGE_SHIFT;
270 	rodata = __pa(__start_rodata);
271 	pfn = rodata >> PAGE_SHIFT;
272 	if (kernel_map_pages_in_pgd(pgd, pfn, rodata, npages, pf)) {
273 		pr_err("Failed to map kernel rodata 1:1\n");
274 		return 1;
275 	}
276 
277 	return 0;
278 }
279 
280 static void __init __map_region(efi_memory_desc_t *md, u64 va)
281 {
282 	unsigned long flags = _PAGE_RW;
283 	unsigned long pfn;
284 	pgd_t *pgd = efi_mm.pgd;
285 
286 	/*
287 	 * EFI_RUNTIME_SERVICES_CODE regions typically cover PE/COFF
288 	 * executable images in memory that consist of both R-X and
289 	 * RW- sections, so we cannot apply read-only or non-exec
290 	 * permissions just yet. However, modern EFI systems provide
291 	 * a memory attributes table that describes those sections
292 	 * with the appropriate restricted permissions, which are
293 	 * applied in efi_runtime_update_mappings() below. All other
294 	 * regions can be mapped non-executable at this point, with
295 	 * the exception of boot services code regions, but those will
296 	 * be unmapped again entirely in efi_free_boot_services().
297 	 */
298 	if (md->type != EFI_BOOT_SERVICES_CODE &&
299 	    md->type != EFI_RUNTIME_SERVICES_CODE)
300 		flags |= _PAGE_NX;
301 
302 	if (!(md->attribute & EFI_MEMORY_WB))
303 		flags |= _PAGE_PCD;
304 
305 	if (sev_active() && md->type != EFI_MEMORY_MAPPED_IO)
306 		flags |= _PAGE_ENC;
307 
308 	pfn = md->phys_addr >> PAGE_SHIFT;
309 	if (kernel_map_pages_in_pgd(pgd, pfn, va, md->num_pages, flags))
310 		pr_warn("Error mapping PA 0x%llx -> VA 0x%llx!\n",
311 			   md->phys_addr, va);
312 }
313 
314 void __init efi_map_region(efi_memory_desc_t *md)
315 {
316 	unsigned long size = md->num_pages << PAGE_SHIFT;
317 	u64 pa = md->phys_addr;
318 
319 	if (efi_have_uv1_memmap())
320 		return old_map_region(md);
321 
322 	/*
323 	 * Make sure the 1:1 mappings are present as a catch-all for b0rked
324 	 * firmware which doesn't update all internal pointers after switching
325 	 * to virtual mode and would otherwise crap on us.
326 	 */
327 	__map_region(md, md->phys_addr);
328 
329 	/*
330 	 * Enforce the 1:1 mapping as the default virtual address when
331 	 * booting in EFI mixed mode, because even though we may be
332 	 * running a 64-bit kernel, the firmware may only be 32-bit.
333 	 */
334 	if (efi_is_mixed()) {
335 		md->virt_addr = md->phys_addr;
336 		return;
337 	}
338 
339 	efi_va -= size;
340 
341 	/* Is PA 2M-aligned? */
342 	if (!(pa & (PMD_SIZE - 1))) {
343 		efi_va &= PMD_MASK;
344 	} else {
345 		u64 pa_offset = pa & (PMD_SIZE - 1);
346 		u64 prev_va = efi_va;
347 
348 		/* get us the same offset within this 2M page */
349 		efi_va = (efi_va & PMD_MASK) + pa_offset;
350 
351 		if (efi_va > prev_va)
352 			efi_va -= PMD_SIZE;
353 	}
354 
355 	if (efi_va < EFI_VA_END) {
356 		pr_warn(FW_WARN "VA address range overflow!\n");
357 		return;
358 	}
359 
360 	/* Do the VA map */
361 	__map_region(md, efi_va);
362 	md->virt_addr = efi_va;
363 }
364 
365 /*
366  * kexec kernel will use efi_map_region_fixed to map efi runtime memory ranges.
367  * md->virt_addr is the original virtual address which had been mapped in kexec
368  * 1st kernel.
369  */
370 void __init efi_map_region_fixed(efi_memory_desc_t *md)
371 {
372 	__map_region(md, md->phys_addr);
373 	__map_region(md, md->virt_addr);
374 }
375 
376 void __init parse_efi_setup(u64 phys_addr, u32 data_len)
377 {
378 	efi_setup = phys_addr + sizeof(struct setup_data);
379 }
380 
381 static int __init efi_update_mappings(efi_memory_desc_t *md, unsigned long pf)
382 {
383 	unsigned long pfn;
384 	pgd_t *pgd = efi_mm.pgd;
385 	int err1, err2;
386 
387 	/* Update the 1:1 mapping */
388 	pfn = md->phys_addr >> PAGE_SHIFT;
389 	err1 = kernel_map_pages_in_pgd(pgd, pfn, md->phys_addr, md->num_pages, pf);
390 	if (err1) {
391 		pr_err("Error while updating 1:1 mapping PA 0x%llx -> VA 0x%llx!\n",
392 			   md->phys_addr, md->virt_addr);
393 	}
394 
395 	err2 = kernel_map_pages_in_pgd(pgd, pfn, md->virt_addr, md->num_pages, pf);
396 	if (err2) {
397 		pr_err("Error while updating VA mapping PA 0x%llx -> VA 0x%llx!\n",
398 			   md->phys_addr, md->virt_addr);
399 	}
400 
401 	return err1 || err2;
402 }
403 
404 static int __init efi_update_mem_attr(struct mm_struct *mm, efi_memory_desc_t *md)
405 {
406 	unsigned long pf = 0;
407 
408 	if (md->attribute & EFI_MEMORY_XP)
409 		pf |= _PAGE_NX;
410 
411 	if (!(md->attribute & EFI_MEMORY_RO))
412 		pf |= _PAGE_RW;
413 
414 	if (sev_active())
415 		pf |= _PAGE_ENC;
416 
417 	return efi_update_mappings(md, pf);
418 }
419 
420 void __init efi_runtime_update_mappings(void)
421 {
422 	efi_memory_desc_t *md;
423 
424 	if (efi_have_uv1_memmap()) {
425 		if (__supported_pte_mask & _PAGE_NX)
426 			runtime_code_page_mkexec();
427 		return;
428 	}
429 
430 	/*
431 	 * Use the EFI Memory Attribute Table for mapping permissions if it
432 	 * exists, since it is intended to supersede EFI_PROPERTIES_TABLE.
433 	 */
434 	if (efi_enabled(EFI_MEM_ATTR)) {
435 		efi_memattr_apply_permissions(NULL, efi_update_mem_attr);
436 		return;
437 	}
438 
439 	/*
440 	 * EFI_MEMORY_ATTRIBUTES_TABLE is intended to replace
441 	 * EFI_PROPERTIES_TABLE. So, use EFI_PROPERTIES_TABLE to update
442 	 * permissions only if EFI_MEMORY_ATTRIBUTES_TABLE is not
443 	 * published by the firmware. Even if we find a buggy implementation of
444 	 * EFI_MEMORY_ATTRIBUTES_TABLE, don't fall back to
445 	 * EFI_PROPERTIES_TABLE, because of the same reason.
446 	 */
447 
448 	if (!efi_enabled(EFI_NX_PE_DATA))
449 		return;
450 
451 	for_each_efi_memory_desc(md) {
452 		unsigned long pf = 0;
453 
454 		if (!(md->attribute & EFI_MEMORY_RUNTIME))
455 			continue;
456 
457 		if (!(md->attribute & EFI_MEMORY_WB))
458 			pf |= _PAGE_PCD;
459 
460 		if ((md->attribute & EFI_MEMORY_XP) ||
461 			(md->type == EFI_RUNTIME_SERVICES_DATA))
462 			pf |= _PAGE_NX;
463 
464 		if (!(md->attribute & EFI_MEMORY_RO) &&
465 			(md->type != EFI_RUNTIME_SERVICES_CODE))
466 			pf |= _PAGE_RW;
467 
468 		if (sev_active())
469 			pf |= _PAGE_ENC;
470 
471 		efi_update_mappings(md, pf);
472 	}
473 }
474 
475 void __init efi_dump_pagetable(void)
476 {
477 #ifdef CONFIG_EFI_PGT_DUMP
478 	if (efi_have_uv1_memmap())
479 		ptdump_walk_pgd_level(NULL, &init_mm);
480 	else
481 		ptdump_walk_pgd_level(NULL, &efi_mm);
482 #endif
483 }
484 
485 /*
486  * Makes the calling thread switch to/from efi_mm context. Can be used
487  * in a kernel thread and user context. Preemption needs to remain disabled
488  * while the EFI-mm is borrowed. mmgrab()/mmdrop() is not used because the mm
489  * can not change under us.
490  * It should be ensured that there are no concurent calls to this function.
491  */
492 void efi_switch_mm(struct mm_struct *mm)
493 {
494 	efi_scratch.prev_mm = current->active_mm;
495 	current->active_mm = mm;
496 	switch_mm(efi_scratch.prev_mm, mm, NULL);
497 }
498 
499 static DEFINE_SPINLOCK(efi_runtime_lock);
500 
501 /*
502  * DS and ES contain user values.  We need to save them.
503  * The 32-bit EFI code needs a valid DS, ES, and SS.  There's no
504  * need to save the old SS: __KERNEL_DS is always acceptable.
505  */
506 #define __efi_thunk(func, ...)						\
507 ({									\
508 	unsigned short __ds, __es;					\
509 	efi_status_t ____s;						\
510 									\
511 	savesegment(ds, __ds);						\
512 	savesegment(es, __es);						\
513 									\
514 	loadsegment(ss, __KERNEL_DS);					\
515 	loadsegment(ds, __KERNEL_DS);					\
516 	loadsegment(es, __KERNEL_DS);					\
517 									\
518 	____s = efi64_thunk(efi.runtime->mixed_mode.func, __VA_ARGS__);	\
519 									\
520 	loadsegment(ds, __ds);						\
521 	loadsegment(es, __es);						\
522 									\
523 	____s ^= (____s & BIT(31)) | (____s & BIT_ULL(31)) << 32;	\
524 	____s;								\
525 })
526 
527 /*
528  * Switch to the EFI page tables early so that we can access the 1:1
529  * runtime services mappings which are not mapped in any other page
530  * tables.
531  *
532  * Also, disable interrupts because the IDT points to 64-bit handlers,
533  * which aren't going to function correctly when we switch to 32-bit.
534  */
535 #define efi_thunk(func...)						\
536 ({									\
537 	efi_status_t __s;						\
538 									\
539 	arch_efi_call_virt_setup();					\
540 									\
541 	__s = __efi_thunk(func);					\
542 									\
543 	arch_efi_call_virt_teardown();					\
544 									\
545 	__s;								\
546 })
547 
548 static efi_status_t __init __no_sanitize_address
549 efi_thunk_set_virtual_address_map(unsigned long memory_map_size,
550 				  unsigned long descriptor_size,
551 				  u32 descriptor_version,
552 				  efi_memory_desc_t *virtual_map)
553 {
554 	efi_status_t status;
555 	unsigned long flags;
556 
557 	efi_sync_low_kernel_mappings();
558 	local_irq_save(flags);
559 
560 	efi_switch_mm(&efi_mm);
561 
562 	status = __efi_thunk(set_virtual_address_map, memory_map_size,
563 			     descriptor_size, descriptor_version, virtual_map);
564 
565 	efi_switch_mm(efi_scratch.prev_mm);
566 	local_irq_restore(flags);
567 
568 	return status;
569 }
570 
571 static efi_status_t efi_thunk_get_time(efi_time_t *tm, efi_time_cap_t *tc)
572 {
573 	return EFI_UNSUPPORTED;
574 }
575 
576 static efi_status_t efi_thunk_set_time(efi_time_t *tm)
577 {
578 	return EFI_UNSUPPORTED;
579 }
580 
581 static efi_status_t
582 efi_thunk_get_wakeup_time(efi_bool_t *enabled, efi_bool_t *pending,
583 			  efi_time_t *tm)
584 {
585 	return EFI_UNSUPPORTED;
586 }
587 
588 static efi_status_t
589 efi_thunk_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm)
590 {
591 	return EFI_UNSUPPORTED;
592 }
593 
594 static unsigned long efi_name_size(efi_char16_t *name)
595 {
596 	return ucs2_strsize(name, EFI_VAR_NAME_LEN) + 1;
597 }
598 
599 static efi_status_t
600 efi_thunk_get_variable(efi_char16_t *name, efi_guid_t *vendor,
601 		       u32 *attr, unsigned long *data_size, void *data)
602 {
603 	u8 buf[24] __aligned(8);
604 	efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
605 	efi_status_t status;
606 	u32 phys_name, phys_vendor, phys_attr;
607 	u32 phys_data_size, phys_data;
608 	unsigned long flags;
609 
610 	spin_lock_irqsave(&efi_runtime_lock, flags);
611 
612 	*vnd = *vendor;
613 
614 	phys_data_size = virt_to_phys_or_null(data_size);
615 	phys_vendor = virt_to_phys_or_null(vnd);
616 	phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
617 	phys_attr = virt_to_phys_or_null(attr);
618 	phys_data = virt_to_phys_or_null_size(data, *data_size);
619 
620 	if (!phys_name || (data && !phys_data))
621 		status = EFI_INVALID_PARAMETER;
622 	else
623 		status = efi_thunk(get_variable, phys_name, phys_vendor,
624 				   phys_attr, phys_data_size, phys_data);
625 
626 	spin_unlock_irqrestore(&efi_runtime_lock, flags);
627 
628 	return status;
629 }
630 
631 static efi_status_t
632 efi_thunk_set_variable(efi_char16_t *name, efi_guid_t *vendor,
633 		       u32 attr, unsigned long data_size, void *data)
634 {
635 	u8 buf[24] __aligned(8);
636 	efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
637 	u32 phys_name, phys_vendor, phys_data;
638 	efi_status_t status;
639 	unsigned long flags;
640 
641 	spin_lock_irqsave(&efi_runtime_lock, flags);
642 
643 	*vnd = *vendor;
644 
645 	phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
646 	phys_vendor = virt_to_phys_or_null(vnd);
647 	phys_data = virt_to_phys_or_null_size(data, data_size);
648 
649 	if (!phys_name || (data && !phys_data))
650 		status = EFI_INVALID_PARAMETER;
651 	else
652 		status = efi_thunk(set_variable, phys_name, phys_vendor,
653 				   attr, data_size, phys_data);
654 
655 	spin_unlock_irqrestore(&efi_runtime_lock, flags);
656 
657 	return status;
658 }
659 
660 static efi_status_t
661 efi_thunk_set_variable_nonblocking(efi_char16_t *name, efi_guid_t *vendor,
662 				   u32 attr, unsigned long data_size,
663 				   void *data)
664 {
665 	u8 buf[24] __aligned(8);
666 	efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
667 	u32 phys_name, phys_vendor, phys_data;
668 	efi_status_t status;
669 	unsigned long flags;
670 
671 	if (!spin_trylock_irqsave(&efi_runtime_lock, flags))
672 		return EFI_NOT_READY;
673 
674 	*vnd = *vendor;
675 
676 	phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
677 	phys_vendor = virt_to_phys_or_null(vnd);
678 	phys_data = virt_to_phys_or_null_size(data, data_size);
679 
680 	if (!phys_name || (data && !phys_data))
681 		status = EFI_INVALID_PARAMETER;
682 	else
683 		status = efi_thunk(set_variable, phys_name, phys_vendor,
684 				   attr, data_size, phys_data);
685 
686 	spin_unlock_irqrestore(&efi_runtime_lock, flags);
687 
688 	return status;
689 }
690 
691 static efi_status_t
692 efi_thunk_get_next_variable(unsigned long *name_size,
693 			    efi_char16_t *name,
694 			    efi_guid_t *vendor)
695 {
696 	u8 buf[24] __aligned(8);
697 	efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
698 	efi_status_t status;
699 	u32 phys_name_size, phys_name, phys_vendor;
700 	unsigned long flags;
701 
702 	spin_lock_irqsave(&efi_runtime_lock, flags);
703 
704 	*vnd = *vendor;
705 
706 	phys_name_size = virt_to_phys_or_null(name_size);
707 	phys_vendor = virt_to_phys_or_null(vnd);
708 	phys_name = virt_to_phys_or_null_size(name, *name_size);
709 
710 	if (!phys_name)
711 		status = EFI_INVALID_PARAMETER;
712 	else
713 		status = efi_thunk(get_next_variable, phys_name_size,
714 				   phys_name, phys_vendor);
715 
716 	spin_unlock_irqrestore(&efi_runtime_lock, flags);
717 
718 	*vendor = *vnd;
719 	return status;
720 }
721 
722 static efi_status_t
723 efi_thunk_get_next_high_mono_count(u32 *count)
724 {
725 	return EFI_UNSUPPORTED;
726 }
727 
728 static void
729 efi_thunk_reset_system(int reset_type, efi_status_t status,
730 		       unsigned long data_size, efi_char16_t *data)
731 {
732 	u32 phys_data;
733 	unsigned long flags;
734 
735 	spin_lock_irqsave(&efi_runtime_lock, flags);
736 
737 	phys_data = virt_to_phys_or_null_size(data, data_size);
738 
739 	efi_thunk(reset_system, reset_type, status, data_size, phys_data);
740 
741 	spin_unlock_irqrestore(&efi_runtime_lock, flags);
742 }
743 
744 static efi_status_t
745 efi_thunk_update_capsule(efi_capsule_header_t **capsules,
746 			 unsigned long count, unsigned long sg_list)
747 {
748 	/*
749 	 * To properly support this function we would need to repackage
750 	 * 'capsules' because the firmware doesn't understand 64-bit
751 	 * pointers.
752 	 */
753 	return EFI_UNSUPPORTED;
754 }
755 
756 static efi_status_t
757 efi_thunk_query_variable_info(u32 attr, u64 *storage_space,
758 			      u64 *remaining_space,
759 			      u64 *max_variable_size)
760 {
761 	efi_status_t status;
762 	u32 phys_storage, phys_remaining, phys_max;
763 	unsigned long flags;
764 
765 	if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
766 		return EFI_UNSUPPORTED;
767 
768 	spin_lock_irqsave(&efi_runtime_lock, flags);
769 
770 	phys_storage = virt_to_phys_or_null(storage_space);
771 	phys_remaining = virt_to_phys_or_null(remaining_space);
772 	phys_max = virt_to_phys_or_null(max_variable_size);
773 
774 	status = efi_thunk(query_variable_info, attr, phys_storage,
775 			   phys_remaining, phys_max);
776 
777 	spin_unlock_irqrestore(&efi_runtime_lock, flags);
778 
779 	return status;
780 }
781 
782 static efi_status_t
783 efi_thunk_query_variable_info_nonblocking(u32 attr, u64 *storage_space,
784 					  u64 *remaining_space,
785 					  u64 *max_variable_size)
786 {
787 	efi_status_t status;
788 	u32 phys_storage, phys_remaining, phys_max;
789 	unsigned long flags;
790 
791 	if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
792 		return EFI_UNSUPPORTED;
793 
794 	if (!spin_trylock_irqsave(&efi_runtime_lock, flags))
795 		return EFI_NOT_READY;
796 
797 	phys_storage = virt_to_phys_or_null(storage_space);
798 	phys_remaining = virt_to_phys_or_null(remaining_space);
799 	phys_max = virt_to_phys_or_null(max_variable_size);
800 
801 	status = efi_thunk(query_variable_info, attr, phys_storage,
802 			   phys_remaining, phys_max);
803 
804 	spin_unlock_irqrestore(&efi_runtime_lock, flags);
805 
806 	return status;
807 }
808 
809 static efi_status_t
810 efi_thunk_query_capsule_caps(efi_capsule_header_t **capsules,
811 			     unsigned long count, u64 *max_size,
812 			     int *reset_type)
813 {
814 	/*
815 	 * To properly support this function we would need to repackage
816 	 * 'capsules' because the firmware doesn't understand 64-bit
817 	 * pointers.
818 	 */
819 	return EFI_UNSUPPORTED;
820 }
821 
822 void __init efi_thunk_runtime_setup(void)
823 {
824 	if (!IS_ENABLED(CONFIG_EFI_MIXED))
825 		return;
826 
827 	efi.get_time = efi_thunk_get_time;
828 	efi.set_time = efi_thunk_set_time;
829 	efi.get_wakeup_time = efi_thunk_get_wakeup_time;
830 	efi.set_wakeup_time = efi_thunk_set_wakeup_time;
831 	efi.get_variable = efi_thunk_get_variable;
832 	efi.get_next_variable = efi_thunk_get_next_variable;
833 	efi.set_variable = efi_thunk_set_variable;
834 	efi.set_variable_nonblocking = efi_thunk_set_variable_nonblocking;
835 	efi.get_next_high_mono_count = efi_thunk_get_next_high_mono_count;
836 	efi.reset_system = efi_thunk_reset_system;
837 	efi.query_variable_info = efi_thunk_query_variable_info;
838 	efi.query_variable_info_nonblocking = efi_thunk_query_variable_info_nonblocking;
839 	efi.update_capsule = efi_thunk_update_capsule;
840 	efi.query_capsule_caps = efi_thunk_query_capsule_caps;
841 }
842 
843 efi_status_t __init __no_sanitize_address
844 efi_set_virtual_address_map(unsigned long memory_map_size,
845 			    unsigned long descriptor_size,
846 			    u32 descriptor_version,
847 			    efi_memory_desc_t *virtual_map,
848 			    unsigned long systab_phys)
849 {
850 	const efi_system_table_t *systab = (efi_system_table_t *)systab_phys;
851 	efi_status_t status;
852 	unsigned long flags;
853 	pgd_t *save_pgd = NULL;
854 
855 	if (efi_is_mixed())
856 		return efi_thunk_set_virtual_address_map(memory_map_size,
857 							 descriptor_size,
858 							 descriptor_version,
859 							 virtual_map);
860 
861 	if (efi_have_uv1_memmap()) {
862 		save_pgd = efi_uv1_memmap_phys_prolog();
863 		if (!save_pgd)
864 			return EFI_ABORTED;
865 	} else {
866 		efi_switch_mm(&efi_mm);
867 	}
868 
869 	kernel_fpu_begin();
870 
871 	/* Disable interrupts around EFI calls: */
872 	local_irq_save(flags);
873 	status = efi_call(efi.runtime->set_virtual_address_map,
874 			  memory_map_size, descriptor_size,
875 			  descriptor_version, virtual_map);
876 	local_irq_restore(flags);
877 
878 	kernel_fpu_end();
879 
880 	/* grab the virtually remapped EFI runtime services table pointer */
881 	efi.runtime = READ_ONCE(systab->runtime);
882 
883 	if (save_pgd)
884 		efi_uv1_memmap_phys_epilog(save_pgd);
885 	else
886 		efi_switch_mm(efi_scratch.prev_mm);
887 
888 	return status;
889 }
890