xref: /linux/arch/x86/platform/efi/efi_64.c (revision e2be04c7f9958dde770eeb8b30e829ca969b37bb)
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/bootmem.h>
27 #include <linux/ioport.h>
28 #include <linux/init.h>
29 #include <linux/mc146818rtc.h>
30 #include <linux/efi.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 
37 #include <asm/setup.h>
38 #include <asm/page.h>
39 #include <asm/e820/api.h>
40 #include <asm/pgtable.h>
41 #include <asm/tlbflush.h>
42 #include <asm/proto.h>
43 #include <asm/efi.h>
44 #include <asm/cacheflush.h>
45 #include <asm/fixmap.h>
46 #include <asm/realmode.h>
47 #include <asm/time.h>
48 #include <asm/pgalloc.h>
49 
50 /*
51  * We allocate runtime services regions top-down, starting from -4G, i.e.
52  * 0xffff_ffff_0000_0000 and limit EFI VA mapping space to 64G.
53  */
54 static u64 efi_va = EFI_VA_START;
55 
56 struct efi_scratch efi_scratch;
57 
58 static void __init early_code_mapping_set_exec(int executable)
59 {
60 	efi_memory_desc_t *md;
61 
62 	if (!(__supported_pte_mask & _PAGE_NX))
63 		return;
64 
65 	/* Make EFI service code area executable */
66 	for_each_efi_memory_desc(md) {
67 		if (md->type == EFI_RUNTIME_SERVICES_CODE ||
68 		    md->type == EFI_BOOT_SERVICES_CODE)
69 			efi_set_executable(md, executable);
70 	}
71 }
72 
73 pgd_t * __init efi_call_phys_prolog(void)
74 {
75 	unsigned long vaddr, addr_pgd, addr_p4d, addr_pud;
76 	pgd_t *save_pgd, *pgd_k, *pgd_efi;
77 	p4d_t *p4d, *p4d_k, *p4d_efi;
78 	pud_t *pud;
79 
80 	int pgd;
81 	int n_pgds, i, j;
82 
83 	if (!efi_enabled(EFI_OLD_MEMMAP)) {
84 		save_pgd = (pgd_t *)__read_cr3();
85 		write_cr3((unsigned long)efi_scratch.efi_pgt);
86 		goto out;
87 	}
88 
89 	early_code_mapping_set_exec(1);
90 
91 	n_pgds = DIV_ROUND_UP((max_pfn << PAGE_SHIFT), PGDIR_SIZE);
92 	save_pgd = kmalloc_array(n_pgds, sizeof(*save_pgd), GFP_KERNEL);
93 
94 	/*
95 	 * Build 1:1 identity mapping for efi=old_map usage. Note that
96 	 * PAGE_OFFSET is PGDIR_SIZE aligned when KASLR is disabled, while
97 	 * it is PUD_SIZE ALIGNED with KASLR enabled. So for a given physical
98 	 * address X, the pud_index(X) != pud_index(__va(X)), we can only copy
99 	 * PUD entry of __va(X) to fill in pud entry of X to build 1:1 mapping.
100 	 * This means here we can only reuse the PMD tables of the direct mapping.
101 	 */
102 	for (pgd = 0; pgd < n_pgds; pgd++) {
103 		addr_pgd = (unsigned long)(pgd * PGDIR_SIZE);
104 		vaddr = (unsigned long)__va(pgd * PGDIR_SIZE);
105 		pgd_efi = pgd_offset_k(addr_pgd);
106 		save_pgd[pgd] = *pgd_efi;
107 
108 		p4d = p4d_alloc(&init_mm, pgd_efi, addr_pgd);
109 		if (!p4d) {
110 			pr_err("Failed to allocate p4d table!\n");
111 			goto out;
112 		}
113 
114 		for (i = 0; i < PTRS_PER_P4D; i++) {
115 			addr_p4d = addr_pgd + i * P4D_SIZE;
116 			p4d_efi = p4d + p4d_index(addr_p4d);
117 
118 			pud = pud_alloc(&init_mm, p4d_efi, addr_p4d);
119 			if (!pud) {
120 				pr_err("Failed to allocate pud table!\n");
121 				goto out;
122 			}
123 
124 			for (j = 0; j < PTRS_PER_PUD; j++) {
125 				addr_pud = addr_p4d + j * PUD_SIZE;
126 
127 				if (addr_pud > (max_pfn << PAGE_SHIFT))
128 					break;
129 
130 				vaddr = (unsigned long)__va(addr_pud);
131 
132 				pgd_k = pgd_offset_k(vaddr);
133 				p4d_k = p4d_offset(pgd_k, vaddr);
134 				pud[j] = *pud_offset(p4d_k, vaddr);
135 			}
136 		}
137 	}
138 out:
139 	__flush_tlb_all();
140 
141 	return save_pgd;
142 }
143 
144 void __init efi_call_phys_epilog(pgd_t *save_pgd)
145 {
146 	/*
147 	 * After the lock is released, the original page table is restored.
148 	 */
149 	int pgd_idx, i;
150 	int nr_pgds;
151 	pgd_t *pgd;
152 	p4d_t *p4d;
153 	pud_t *pud;
154 
155 	if (!efi_enabled(EFI_OLD_MEMMAP)) {
156 		write_cr3((unsigned long)save_pgd);
157 		__flush_tlb_all();
158 		return;
159 	}
160 
161 	nr_pgds = DIV_ROUND_UP((max_pfn << PAGE_SHIFT) , PGDIR_SIZE);
162 
163 	for (pgd_idx = 0; pgd_idx < nr_pgds; pgd_idx++) {
164 		pgd = pgd_offset_k(pgd_idx * PGDIR_SIZE);
165 		set_pgd(pgd_offset_k(pgd_idx * PGDIR_SIZE), save_pgd[pgd_idx]);
166 
167 		if (!(pgd_val(*pgd) & _PAGE_PRESENT))
168 			continue;
169 
170 		for (i = 0; i < PTRS_PER_P4D; i++) {
171 			p4d = p4d_offset(pgd,
172 					 pgd_idx * PGDIR_SIZE + i * P4D_SIZE);
173 
174 			if (!(p4d_val(*p4d) & _PAGE_PRESENT))
175 				continue;
176 
177 			pud = (pud_t *)p4d_page_vaddr(*p4d);
178 			pud_free(&init_mm, pud);
179 		}
180 
181 		p4d = (p4d_t *)pgd_page_vaddr(*pgd);
182 		p4d_free(&init_mm, p4d);
183 	}
184 
185 	kfree(save_pgd);
186 
187 	__flush_tlb_all();
188 	early_code_mapping_set_exec(0);
189 }
190 
191 static pgd_t *efi_pgd;
192 
193 /*
194  * We need our own copy of the higher levels of the page tables
195  * because we want to avoid inserting EFI region mappings (EFI_VA_END
196  * to EFI_VA_START) into the standard kernel page tables. Everything
197  * else can be shared, see efi_sync_low_kernel_mappings().
198  */
199 int __init efi_alloc_page_tables(void)
200 {
201 	pgd_t *pgd;
202 	p4d_t *p4d;
203 	pud_t *pud;
204 	gfp_t gfp_mask;
205 
206 	if (efi_enabled(EFI_OLD_MEMMAP))
207 		return 0;
208 
209 	gfp_mask = GFP_KERNEL | __GFP_NOTRACK | __GFP_ZERO;
210 	efi_pgd = (pgd_t *)__get_free_page(gfp_mask);
211 	if (!efi_pgd)
212 		return -ENOMEM;
213 
214 	pgd = efi_pgd + pgd_index(EFI_VA_END);
215 	p4d = p4d_alloc(&init_mm, pgd, EFI_VA_END);
216 	if (!p4d) {
217 		free_page((unsigned long)efi_pgd);
218 		return -ENOMEM;
219 	}
220 
221 	pud = pud_alloc(&init_mm, p4d, EFI_VA_END);
222 	if (!pud) {
223 		if (CONFIG_PGTABLE_LEVELS > 4)
224 			free_page((unsigned long) pgd_page_vaddr(*pgd));
225 		free_page((unsigned long)efi_pgd);
226 		return -ENOMEM;
227 	}
228 
229 	return 0;
230 }
231 
232 /*
233  * Add low kernel mappings for passing arguments to EFI functions.
234  */
235 void efi_sync_low_kernel_mappings(void)
236 {
237 	unsigned num_entries;
238 	pgd_t *pgd_k, *pgd_efi;
239 	p4d_t *p4d_k, *p4d_efi;
240 	pud_t *pud_k, *pud_efi;
241 
242 	if (efi_enabled(EFI_OLD_MEMMAP))
243 		return;
244 
245 	/*
246 	 * We can share all PGD entries apart from the one entry that
247 	 * covers the EFI runtime mapping space.
248 	 *
249 	 * Make sure the EFI runtime region mappings are guaranteed to
250 	 * only span a single PGD entry and that the entry also maps
251 	 * other important kernel regions.
252 	 */
253 	BUILD_BUG_ON(pgd_index(EFI_VA_END) != pgd_index(MODULES_END));
254 	BUILD_BUG_ON((EFI_VA_START & PGDIR_MASK) !=
255 			(EFI_VA_END & PGDIR_MASK));
256 
257 	pgd_efi = efi_pgd + pgd_index(PAGE_OFFSET);
258 	pgd_k = pgd_offset_k(PAGE_OFFSET);
259 
260 	num_entries = pgd_index(EFI_VA_END) - pgd_index(PAGE_OFFSET);
261 	memcpy(pgd_efi, pgd_k, sizeof(pgd_t) * num_entries);
262 
263 	/*
264 	 * As with PGDs, we share all P4D entries apart from the one entry
265 	 * that covers the EFI runtime mapping space.
266 	 */
267 	BUILD_BUG_ON(p4d_index(EFI_VA_END) != p4d_index(MODULES_END));
268 	BUILD_BUG_ON((EFI_VA_START & P4D_MASK) != (EFI_VA_END & P4D_MASK));
269 
270 	pgd_efi = efi_pgd + pgd_index(EFI_VA_END);
271 	pgd_k = pgd_offset_k(EFI_VA_END);
272 	p4d_efi = p4d_offset(pgd_efi, 0);
273 	p4d_k = p4d_offset(pgd_k, 0);
274 
275 	num_entries = p4d_index(EFI_VA_END);
276 	memcpy(p4d_efi, p4d_k, sizeof(p4d_t) * num_entries);
277 
278 	/*
279 	 * We share all the PUD entries apart from those that map the
280 	 * EFI regions. Copy around them.
281 	 */
282 	BUILD_BUG_ON((EFI_VA_START & ~PUD_MASK) != 0);
283 	BUILD_BUG_ON((EFI_VA_END & ~PUD_MASK) != 0);
284 
285 	p4d_efi = p4d_offset(pgd_efi, EFI_VA_END);
286 	p4d_k = p4d_offset(pgd_k, EFI_VA_END);
287 	pud_efi = pud_offset(p4d_efi, 0);
288 	pud_k = pud_offset(p4d_k, 0);
289 
290 	num_entries = pud_index(EFI_VA_END);
291 	memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
292 
293 	pud_efi = pud_offset(p4d_efi, EFI_VA_START);
294 	pud_k = pud_offset(p4d_k, EFI_VA_START);
295 
296 	num_entries = PTRS_PER_PUD - pud_index(EFI_VA_START);
297 	memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
298 }
299 
300 /*
301  * Wrapper for slow_virt_to_phys() that handles NULL addresses.
302  */
303 static inline phys_addr_t
304 virt_to_phys_or_null_size(void *va, unsigned long size)
305 {
306 	bool bad_size;
307 
308 	if (!va)
309 		return 0;
310 
311 	if (virt_addr_valid(va))
312 		return virt_to_phys(va);
313 
314 	/*
315 	 * A fully aligned variable on the stack is guaranteed not to
316 	 * cross a page bounary. Try to catch strings on the stack by
317 	 * checking that 'size' is a power of two.
318 	 */
319 	bad_size = size > PAGE_SIZE || !is_power_of_2(size);
320 
321 	WARN_ON(!IS_ALIGNED((unsigned long)va, size) || bad_size);
322 
323 	return slow_virt_to_phys(va);
324 }
325 
326 #define virt_to_phys_or_null(addr)				\
327 	virt_to_phys_or_null_size((addr), sizeof(*(addr)))
328 
329 int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)
330 {
331 	unsigned long pfn, text, pf;
332 	struct page *page;
333 	unsigned npages;
334 	pgd_t *pgd;
335 
336 	if (efi_enabled(EFI_OLD_MEMMAP))
337 		return 0;
338 
339 	/*
340 	 * Since the PGD is encrypted, set the encryption mask so that when
341 	 * this value is loaded into cr3 the PGD will be decrypted during
342 	 * the pagetable walk.
343 	 */
344 	efi_scratch.efi_pgt = (pgd_t *)__sme_pa(efi_pgd);
345 	pgd = efi_pgd;
346 
347 	/*
348 	 * It can happen that the physical address of new_memmap lands in memory
349 	 * which is not mapped in the EFI page table. Therefore we need to go
350 	 * and ident-map those pages containing the map before calling
351 	 * phys_efi_set_virtual_address_map().
352 	 */
353 	pfn = pa_memmap >> PAGE_SHIFT;
354 	pf = _PAGE_NX | _PAGE_RW | _PAGE_ENC;
355 	if (kernel_map_pages_in_pgd(pgd, pfn, pa_memmap, num_pages, pf)) {
356 		pr_err("Error ident-mapping new memmap (0x%lx)!\n", pa_memmap);
357 		return 1;
358 	}
359 
360 	efi_scratch.use_pgd = true;
361 
362 	/*
363 	 * Certain firmware versions are way too sentimential and still believe
364 	 * they are exclusive and unquestionable owners of the first physical page,
365 	 * even though they explicitly mark it as EFI_CONVENTIONAL_MEMORY
366 	 * (but then write-access it later during SetVirtualAddressMap()).
367 	 *
368 	 * Create a 1:1 mapping for this page, to avoid triple faults during early
369 	 * boot with such firmware. We are free to hand this page to the BIOS,
370 	 * as trim_bios_range() will reserve the first page and isolate it away
371 	 * from memory allocators anyway.
372 	 */
373 	if (kernel_map_pages_in_pgd(pgd, 0x0, 0x0, 1, _PAGE_RW)) {
374 		pr_err("Failed to create 1:1 mapping for the first page!\n");
375 		return 1;
376 	}
377 
378 	/*
379 	 * When making calls to the firmware everything needs to be 1:1
380 	 * mapped and addressable with 32-bit pointers. Map the kernel
381 	 * text and allocate a new stack because we can't rely on the
382 	 * stack pointer being < 4GB.
383 	 */
384 	if (!IS_ENABLED(CONFIG_EFI_MIXED) || efi_is_native())
385 		return 0;
386 
387 	page = alloc_page(GFP_KERNEL|__GFP_DMA32);
388 	if (!page)
389 		panic("Unable to allocate EFI runtime stack < 4GB\n");
390 
391 	efi_scratch.phys_stack = virt_to_phys(page_address(page));
392 	efi_scratch.phys_stack += PAGE_SIZE; /* stack grows down */
393 
394 	npages = (_etext - _text) >> PAGE_SHIFT;
395 	text = __pa(_text);
396 	pfn = text >> PAGE_SHIFT;
397 
398 	pf = _PAGE_RW | _PAGE_ENC;
399 	if (kernel_map_pages_in_pgd(pgd, pfn, text, npages, pf)) {
400 		pr_err("Failed to map kernel text 1:1\n");
401 		return 1;
402 	}
403 
404 	return 0;
405 }
406 
407 static void __init __map_region(efi_memory_desc_t *md, u64 va)
408 {
409 	unsigned long flags = _PAGE_RW;
410 	unsigned long pfn;
411 	pgd_t *pgd = efi_pgd;
412 
413 	if (!(md->attribute & EFI_MEMORY_WB))
414 		flags |= _PAGE_PCD;
415 
416 	pfn = md->phys_addr >> PAGE_SHIFT;
417 	if (kernel_map_pages_in_pgd(pgd, pfn, va, md->num_pages, flags))
418 		pr_warn("Error mapping PA 0x%llx -> VA 0x%llx!\n",
419 			   md->phys_addr, va);
420 }
421 
422 void __init efi_map_region(efi_memory_desc_t *md)
423 {
424 	unsigned long size = md->num_pages << PAGE_SHIFT;
425 	u64 pa = md->phys_addr;
426 
427 	if (efi_enabled(EFI_OLD_MEMMAP))
428 		return old_map_region(md);
429 
430 	/*
431 	 * Make sure the 1:1 mappings are present as a catch-all for b0rked
432 	 * firmware which doesn't update all internal pointers after switching
433 	 * to virtual mode and would otherwise crap on us.
434 	 */
435 	__map_region(md, md->phys_addr);
436 
437 	/*
438 	 * Enforce the 1:1 mapping as the default virtual address when
439 	 * booting in EFI mixed mode, because even though we may be
440 	 * running a 64-bit kernel, the firmware may only be 32-bit.
441 	 */
442 	if (!efi_is_native () && IS_ENABLED(CONFIG_EFI_MIXED)) {
443 		md->virt_addr = md->phys_addr;
444 		return;
445 	}
446 
447 	efi_va -= size;
448 
449 	/* Is PA 2M-aligned? */
450 	if (!(pa & (PMD_SIZE - 1))) {
451 		efi_va &= PMD_MASK;
452 	} else {
453 		u64 pa_offset = pa & (PMD_SIZE - 1);
454 		u64 prev_va = efi_va;
455 
456 		/* get us the same offset within this 2M page */
457 		efi_va = (efi_va & PMD_MASK) + pa_offset;
458 
459 		if (efi_va > prev_va)
460 			efi_va -= PMD_SIZE;
461 	}
462 
463 	if (efi_va < EFI_VA_END) {
464 		pr_warn(FW_WARN "VA address range overflow!\n");
465 		return;
466 	}
467 
468 	/* Do the VA map */
469 	__map_region(md, efi_va);
470 	md->virt_addr = efi_va;
471 }
472 
473 /*
474  * kexec kernel will use efi_map_region_fixed to map efi runtime memory ranges.
475  * md->virt_addr is the original virtual address which had been mapped in kexec
476  * 1st kernel.
477  */
478 void __init efi_map_region_fixed(efi_memory_desc_t *md)
479 {
480 	__map_region(md, md->phys_addr);
481 	__map_region(md, md->virt_addr);
482 }
483 
484 void __iomem *__init efi_ioremap(unsigned long phys_addr, unsigned long size,
485 				 u32 type, u64 attribute)
486 {
487 	unsigned long last_map_pfn;
488 
489 	if (type == EFI_MEMORY_MAPPED_IO)
490 		return ioremap(phys_addr, size);
491 
492 	last_map_pfn = init_memory_mapping(phys_addr, phys_addr + size);
493 	if ((last_map_pfn << PAGE_SHIFT) < phys_addr + size) {
494 		unsigned long top = last_map_pfn << PAGE_SHIFT;
495 		efi_ioremap(top, size - (top - phys_addr), type, attribute);
496 	}
497 
498 	if (!(attribute & EFI_MEMORY_WB))
499 		efi_memory_uc((u64)(unsigned long)__va(phys_addr), size);
500 
501 	return (void __iomem *)__va(phys_addr);
502 }
503 
504 void __init parse_efi_setup(u64 phys_addr, u32 data_len)
505 {
506 	efi_setup = phys_addr + sizeof(struct setup_data);
507 }
508 
509 static int __init efi_update_mappings(efi_memory_desc_t *md, unsigned long pf)
510 {
511 	unsigned long pfn;
512 	pgd_t *pgd = efi_pgd;
513 	int err1, err2;
514 
515 	/* Update the 1:1 mapping */
516 	pfn = md->phys_addr >> PAGE_SHIFT;
517 	err1 = kernel_map_pages_in_pgd(pgd, pfn, md->phys_addr, md->num_pages, pf);
518 	if (err1) {
519 		pr_err("Error while updating 1:1 mapping PA 0x%llx -> VA 0x%llx!\n",
520 			   md->phys_addr, md->virt_addr);
521 	}
522 
523 	err2 = kernel_map_pages_in_pgd(pgd, pfn, md->virt_addr, md->num_pages, pf);
524 	if (err2) {
525 		pr_err("Error while updating VA mapping PA 0x%llx -> VA 0x%llx!\n",
526 			   md->phys_addr, md->virt_addr);
527 	}
528 
529 	return err1 || err2;
530 }
531 
532 static int __init efi_update_mem_attr(struct mm_struct *mm, efi_memory_desc_t *md)
533 {
534 	unsigned long pf = 0;
535 
536 	if (md->attribute & EFI_MEMORY_XP)
537 		pf |= _PAGE_NX;
538 
539 	if (!(md->attribute & EFI_MEMORY_RO))
540 		pf |= _PAGE_RW;
541 
542 	return efi_update_mappings(md, pf);
543 }
544 
545 void __init efi_runtime_update_mappings(void)
546 {
547 	efi_memory_desc_t *md;
548 
549 	if (efi_enabled(EFI_OLD_MEMMAP)) {
550 		if (__supported_pte_mask & _PAGE_NX)
551 			runtime_code_page_mkexec();
552 		return;
553 	}
554 
555 	/*
556 	 * Use the EFI Memory Attribute Table for mapping permissions if it
557 	 * exists, since it is intended to supersede EFI_PROPERTIES_TABLE.
558 	 */
559 	if (efi_enabled(EFI_MEM_ATTR)) {
560 		efi_memattr_apply_permissions(NULL, efi_update_mem_attr);
561 		return;
562 	}
563 
564 	/*
565 	 * EFI_MEMORY_ATTRIBUTES_TABLE is intended to replace
566 	 * EFI_PROPERTIES_TABLE. So, use EFI_PROPERTIES_TABLE to update
567 	 * permissions only if EFI_MEMORY_ATTRIBUTES_TABLE is not
568 	 * published by the firmware. Even if we find a buggy implementation of
569 	 * EFI_MEMORY_ATTRIBUTES_TABLE, don't fall back to
570 	 * EFI_PROPERTIES_TABLE, because of the same reason.
571 	 */
572 
573 	if (!efi_enabled(EFI_NX_PE_DATA))
574 		return;
575 
576 	for_each_efi_memory_desc(md) {
577 		unsigned long pf = 0;
578 
579 		if (!(md->attribute & EFI_MEMORY_RUNTIME))
580 			continue;
581 
582 		if (!(md->attribute & EFI_MEMORY_WB))
583 			pf |= _PAGE_PCD;
584 
585 		if ((md->attribute & EFI_MEMORY_XP) ||
586 			(md->type == EFI_RUNTIME_SERVICES_DATA))
587 			pf |= _PAGE_NX;
588 
589 		if (!(md->attribute & EFI_MEMORY_RO) &&
590 			(md->type != EFI_RUNTIME_SERVICES_CODE))
591 			pf |= _PAGE_RW;
592 
593 		efi_update_mappings(md, pf);
594 	}
595 }
596 
597 void __init efi_dump_pagetable(void)
598 {
599 #ifdef CONFIG_EFI_PGT_DUMP
600 	if (efi_enabled(EFI_OLD_MEMMAP))
601 		ptdump_walk_pgd_level(NULL, swapper_pg_dir);
602 	else
603 		ptdump_walk_pgd_level(NULL, efi_pgd);
604 #endif
605 }
606 
607 #ifdef CONFIG_EFI_MIXED
608 extern efi_status_t efi64_thunk(u32, ...);
609 
610 #define runtime_service32(func)						 \
611 ({									 \
612 	u32 table = (u32)(unsigned long)efi.systab;			 \
613 	u32 *rt, *___f;							 \
614 									 \
615 	rt = (u32 *)(table + offsetof(efi_system_table_32_t, runtime));	 \
616 	___f = (u32 *)(*rt + offsetof(efi_runtime_services_32_t, func)); \
617 	*___f;								 \
618 })
619 
620 /*
621  * Switch to the EFI page tables early so that we can access the 1:1
622  * runtime services mappings which are not mapped in any other page
623  * tables. This function must be called before runtime_service32().
624  *
625  * Also, disable interrupts because the IDT points to 64-bit handlers,
626  * which aren't going to function correctly when we switch to 32-bit.
627  */
628 #define efi_thunk(f, ...)						\
629 ({									\
630 	efi_status_t __s;						\
631 	unsigned long __flags;						\
632 	u32 __func;							\
633 									\
634 	local_irq_save(__flags);					\
635 	arch_efi_call_virt_setup();					\
636 									\
637 	__func = runtime_service32(f);					\
638 	__s = efi64_thunk(__func, __VA_ARGS__);				\
639 									\
640 	arch_efi_call_virt_teardown();					\
641 	local_irq_restore(__flags);					\
642 									\
643 	__s;								\
644 })
645 
646 efi_status_t efi_thunk_set_virtual_address_map(
647 	void *phys_set_virtual_address_map,
648 	unsigned long memory_map_size,
649 	unsigned long descriptor_size,
650 	u32 descriptor_version,
651 	efi_memory_desc_t *virtual_map)
652 {
653 	efi_status_t status;
654 	unsigned long flags;
655 	u32 func;
656 
657 	efi_sync_low_kernel_mappings();
658 	local_irq_save(flags);
659 
660 	efi_scratch.prev_cr3 = __read_cr3();
661 	write_cr3((unsigned long)efi_scratch.efi_pgt);
662 	__flush_tlb_all();
663 
664 	func = (u32)(unsigned long)phys_set_virtual_address_map;
665 	status = efi64_thunk(func, memory_map_size, descriptor_size,
666 			     descriptor_version, virtual_map);
667 
668 	write_cr3(efi_scratch.prev_cr3);
669 	__flush_tlb_all();
670 	local_irq_restore(flags);
671 
672 	return status;
673 }
674 
675 static efi_status_t efi_thunk_get_time(efi_time_t *tm, efi_time_cap_t *tc)
676 {
677 	efi_status_t status;
678 	u32 phys_tm, phys_tc;
679 
680 	spin_lock(&rtc_lock);
681 
682 	phys_tm = virt_to_phys_or_null(tm);
683 	phys_tc = virt_to_phys_or_null(tc);
684 
685 	status = efi_thunk(get_time, phys_tm, phys_tc);
686 
687 	spin_unlock(&rtc_lock);
688 
689 	return status;
690 }
691 
692 static efi_status_t efi_thunk_set_time(efi_time_t *tm)
693 {
694 	efi_status_t status;
695 	u32 phys_tm;
696 
697 	spin_lock(&rtc_lock);
698 
699 	phys_tm = virt_to_phys_or_null(tm);
700 
701 	status = efi_thunk(set_time, phys_tm);
702 
703 	spin_unlock(&rtc_lock);
704 
705 	return status;
706 }
707 
708 static efi_status_t
709 efi_thunk_get_wakeup_time(efi_bool_t *enabled, efi_bool_t *pending,
710 			  efi_time_t *tm)
711 {
712 	efi_status_t status;
713 	u32 phys_enabled, phys_pending, phys_tm;
714 
715 	spin_lock(&rtc_lock);
716 
717 	phys_enabled = virt_to_phys_or_null(enabled);
718 	phys_pending = virt_to_phys_or_null(pending);
719 	phys_tm = virt_to_phys_or_null(tm);
720 
721 	status = efi_thunk(get_wakeup_time, phys_enabled,
722 			     phys_pending, phys_tm);
723 
724 	spin_unlock(&rtc_lock);
725 
726 	return status;
727 }
728 
729 static efi_status_t
730 efi_thunk_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm)
731 {
732 	efi_status_t status;
733 	u32 phys_tm;
734 
735 	spin_lock(&rtc_lock);
736 
737 	phys_tm = virt_to_phys_or_null(tm);
738 
739 	status = efi_thunk(set_wakeup_time, enabled, phys_tm);
740 
741 	spin_unlock(&rtc_lock);
742 
743 	return status;
744 }
745 
746 static unsigned long efi_name_size(efi_char16_t *name)
747 {
748 	return ucs2_strsize(name, EFI_VAR_NAME_LEN) + 1;
749 }
750 
751 static efi_status_t
752 efi_thunk_get_variable(efi_char16_t *name, efi_guid_t *vendor,
753 		       u32 *attr, unsigned long *data_size, void *data)
754 {
755 	efi_status_t status;
756 	u32 phys_name, phys_vendor, phys_attr;
757 	u32 phys_data_size, phys_data;
758 
759 	phys_data_size = virt_to_phys_or_null(data_size);
760 	phys_vendor = virt_to_phys_or_null(vendor);
761 	phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
762 	phys_attr = virt_to_phys_or_null(attr);
763 	phys_data = virt_to_phys_or_null_size(data, *data_size);
764 
765 	status = efi_thunk(get_variable, phys_name, phys_vendor,
766 			   phys_attr, phys_data_size, phys_data);
767 
768 	return status;
769 }
770 
771 static efi_status_t
772 efi_thunk_set_variable(efi_char16_t *name, efi_guid_t *vendor,
773 		       u32 attr, unsigned long data_size, void *data)
774 {
775 	u32 phys_name, phys_vendor, phys_data;
776 	efi_status_t status;
777 
778 	phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
779 	phys_vendor = virt_to_phys_or_null(vendor);
780 	phys_data = virt_to_phys_or_null_size(data, data_size);
781 
782 	/* If data_size is > sizeof(u32) we've got problems */
783 	status = efi_thunk(set_variable, phys_name, phys_vendor,
784 			   attr, data_size, phys_data);
785 
786 	return status;
787 }
788 
789 static efi_status_t
790 efi_thunk_get_next_variable(unsigned long *name_size,
791 			    efi_char16_t *name,
792 			    efi_guid_t *vendor)
793 {
794 	efi_status_t status;
795 	u32 phys_name_size, phys_name, phys_vendor;
796 
797 	phys_name_size = virt_to_phys_or_null(name_size);
798 	phys_vendor = virt_to_phys_or_null(vendor);
799 	phys_name = virt_to_phys_or_null_size(name, *name_size);
800 
801 	status = efi_thunk(get_next_variable, phys_name_size,
802 			   phys_name, phys_vendor);
803 
804 	return status;
805 }
806 
807 static efi_status_t
808 efi_thunk_get_next_high_mono_count(u32 *count)
809 {
810 	efi_status_t status;
811 	u32 phys_count;
812 
813 	phys_count = virt_to_phys_or_null(count);
814 	status = efi_thunk(get_next_high_mono_count, phys_count);
815 
816 	return status;
817 }
818 
819 static void
820 efi_thunk_reset_system(int reset_type, efi_status_t status,
821 		       unsigned long data_size, efi_char16_t *data)
822 {
823 	u32 phys_data;
824 
825 	phys_data = virt_to_phys_or_null_size(data, data_size);
826 
827 	efi_thunk(reset_system, reset_type, status, data_size, phys_data);
828 }
829 
830 static efi_status_t
831 efi_thunk_update_capsule(efi_capsule_header_t **capsules,
832 			 unsigned long count, unsigned long sg_list)
833 {
834 	/*
835 	 * To properly support this function we would need to repackage
836 	 * 'capsules' because the firmware doesn't understand 64-bit
837 	 * pointers.
838 	 */
839 	return EFI_UNSUPPORTED;
840 }
841 
842 static efi_status_t
843 efi_thunk_query_variable_info(u32 attr, u64 *storage_space,
844 			      u64 *remaining_space,
845 			      u64 *max_variable_size)
846 {
847 	efi_status_t status;
848 	u32 phys_storage, phys_remaining, phys_max;
849 
850 	if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
851 		return EFI_UNSUPPORTED;
852 
853 	phys_storage = virt_to_phys_or_null(storage_space);
854 	phys_remaining = virt_to_phys_or_null(remaining_space);
855 	phys_max = virt_to_phys_or_null(max_variable_size);
856 
857 	status = efi_thunk(query_variable_info, attr, phys_storage,
858 			   phys_remaining, phys_max);
859 
860 	return status;
861 }
862 
863 static efi_status_t
864 efi_thunk_query_capsule_caps(efi_capsule_header_t **capsules,
865 			     unsigned long count, u64 *max_size,
866 			     int *reset_type)
867 {
868 	/*
869 	 * To properly support this function we would need to repackage
870 	 * 'capsules' because the firmware doesn't understand 64-bit
871 	 * pointers.
872 	 */
873 	return EFI_UNSUPPORTED;
874 }
875 
876 void efi_thunk_runtime_setup(void)
877 {
878 	efi.get_time = efi_thunk_get_time;
879 	efi.set_time = efi_thunk_set_time;
880 	efi.get_wakeup_time = efi_thunk_get_wakeup_time;
881 	efi.set_wakeup_time = efi_thunk_set_wakeup_time;
882 	efi.get_variable = efi_thunk_get_variable;
883 	efi.get_next_variable = efi_thunk_get_next_variable;
884 	efi.set_variable = efi_thunk_set_variable;
885 	efi.get_next_high_mono_count = efi_thunk_get_next_high_mono_count;
886 	efi.reset_system = efi_thunk_reset_system;
887 	efi.query_variable_info = efi_thunk_query_variable_info;
888 	efi.update_capsule = efi_thunk_update_capsule;
889 	efi.query_capsule_caps = efi_thunk_query_capsule_caps;
890 }
891 #endif /* CONFIG_EFI_MIXED */
892