xref: /linux/arch/x86/platform/efi/efi_64.c (revision 140eb5227767c6754742020a16d2691222b9c19b)
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 #include <linux/mem_encrypt.h>
37 
38 #include <asm/setup.h>
39 #include <asm/page.h>
40 #include <asm/e820/api.h>
41 #include <asm/pgtable.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 
51 /*
52  * We allocate runtime services regions top-down, starting from -4G, i.e.
53  * 0xffff_ffff_0000_0000 and limit EFI VA mapping space to 64G.
54  */
55 static u64 efi_va = EFI_VA_START;
56 
57 struct efi_scratch efi_scratch;
58 
59 static void __init early_code_mapping_set_exec(int executable)
60 {
61 	efi_memory_desc_t *md;
62 
63 	if (!(__supported_pte_mask & _PAGE_NX))
64 		return;
65 
66 	/* Make EFI service code area executable */
67 	for_each_efi_memory_desc(md) {
68 		if (md->type == EFI_RUNTIME_SERVICES_CODE ||
69 		    md->type == EFI_BOOT_SERVICES_CODE)
70 			efi_set_executable(md, executable);
71 	}
72 }
73 
74 pgd_t * __init efi_call_phys_prolog(void)
75 {
76 	unsigned long vaddr, addr_pgd, addr_p4d, addr_pud;
77 	pgd_t *save_pgd, *pgd_k, *pgd_efi;
78 	p4d_t *p4d, *p4d_k, *p4d_efi;
79 	pud_t *pud;
80 
81 	int pgd;
82 	int n_pgds, i, j;
83 
84 	if (!efi_enabled(EFI_OLD_MEMMAP)) {
85 		save_pgd = (pgd_t *)__read_cr3();
86 		write_cr3((unsigned long)efi_scratch.efi_pgt);
87 		goto out;
88 	}
89 
90 	early_code_mapping_set_exec(1);
91 
92 	n_pgds = DIV_ROUND_UP((max_pfn << PAGE_SHIFT), PGDIR_SIZE);
93 	save_pgd = kmalloc_array(n_pgds, sizeof(*save_pgd), GFP_KERNEL);
94 
95 	/*
96 	 * Build 1:1 identity mapping for efi=old_map usage. Note that
97 	 * PAGE_OFFSET is PGDIR_SIZE aligned when KASLR is disabled, while
98 	 * it is PUD_SIZE ALIGNED with KASLR enabled. So for a given physical
99 	 * address X, the pud_index(X) != pud_index(__va(X)), we can only copy
100 	 * PUD entry of __va(X) to fill in pud entry of X to build 1:1 mapping.
101 	 * This means here we can only reuse the PMD tables of the direct mapping.
102 	 */
103 	for (pgd = 0; pgd < n_pgds; pgd++) {
104 		addr_pgd = (unsigned long)(pgd * PGDIR_SIZE);
105 		vaddr = (unsigned long)__va(pgd * PGDIR_SIZE);
106 		pgd_efi = pgd_offset_k(addr_pgd);
107 		save_pgd[pgd] = *pgd_efi;
108 
109 		p4d = p4d_alloc(&init_mm, pgd_efi, addr_pgd);
110 		if (!p4d) {
111 			pr_err("Failed to allocate p4d table!\n");
112 			goto out;
113 		}
114 
115 		for (i = 0; i < PTRS_PER_P4D; i++) {
116 			addr_p4d = addr_pgd + i * P4D_SIZE;
117 			p4d_efi = p4d + p4d_index(addr_p4d);
118 
119 			pud = pud_alloc(&init_mm, p4d_efi, addr_p4d);
120 			if (!pud) {
121 				pr_err("Failed to allocate pud table!\n");
122 				goto out;
123 			}
124 
125 			for (j = 0; j < PTRS_PER_PUD; j++) {
126 				addr_pud = addr_p4d + j * PUD_SIZE;
127 
128 				if (addr_pud > (max_pfn << PAGE_SHIFT))
129 					break;
130 
131 				vaddr = (unsigned long)__va(addr_pud);
132 
133 				pgd_k = pgd_offset_k(vaddr);
134 				p4d_k = p4d_offset(pgd_k, vaddr);
135 				pud[j] = *pud_offset(p4d_k, vaddr);
136 			}
137 		}
138 		pgd_offset_k(pgd * PGDIR_SIZE)->pgd &= ~_PAGE_NX;
139 	}
140 
141 out:
142 	__flush_tlb_all();
143 
144 	return save_pgd;
145 }
146 
147 void __init efi_call_phys_epilog(pgd_t *save_pgd)
148 {
149 	/*
150 	 * After the lock is released, the original page table is restored.
151 	 */
152 	int pgd_idx, i;
153 	int nr_pgds;
154 	pgd_t *pgd;
155 	p4d_t *p4d;
156 	pud_t *pud;
157 
158 	if (!efi_enabled(EFI_OLD_MEMMAP)) {
159 		write_cr3((unsigned long)save_pgd);
160 		__flush_tlb_all();
161 		return;
162 	}
163 
164 	nr_pgds = DIV_ROUND_UP((max_pfn << PAGE_SHIFT) , PGDIR_SIZE);
165 
166 	for (pgd_idx = 0; pgd_idx < nr_pgds; pgd_idx++) {
167 		pgd = pgd_offset_k(pgd_idx * PGDIR_SIZE);
168 		set_pgd(pgd_offset_k(pgd_idx * PGDIR_SIZE), save_pgd[pgd_idx]);
169 
170 		if (!(pgd_val(*pgd) & _PAGE_PRESENT))
171 			continue;
172 
173 		for (i = 0; i < PTRS_PER_P4D; i++) {
174 			p4d = p4d_offset(pgd,
175 					 pgd_idx * PGDIR_SIZE + i * P4D_SIZE);
176 
177 			if (!(p4d_val(*p4d) & _PAGE_PRESENT))
178 				continue;
179 
180 			pud = (pud_t *)p4d_page_vaddr(*p4d);
181 			pud_free(&init_mm, pud);
182 		}
183 
184 		p4d = (p4d_t *)pgd_page_vaddr(*pgd);
185 		p4d_free(&init_mm, p4d);
186 	}
187 
188 	kfree(save_pgd);
189 
190 	__flush_tlb_all();
191 	early_code_mapping_set_exec(0);
192 }
193 
194 static pgd_t *efi_pgd;
195 
196 /*
197  * We need our own copy of the higher levels of the page tables
198  * because we want to avoid inserting EFI region mappings (EFI_VA_END
199  * to EFI_VA_START) into the standard kernel page tables. Everything
200  * else can be shared, see efi_sync_low_kernel_mappings().
201  *
202  * We don't want the pgd on the pgd_list and cannot use pgd_alloc() for the
203  * allocation.
204  */
205 int __init efi_alloc_page_tables(void)
206 {
207 	pgd_t *pgd;
208 	p4d_t *p4d;
209 	pud_t *pud;
210 	gfp_t gfp_mask;
211 
212 	if (efi_enabled(EFI_OLD_MEMMAP))
213 		return 0;
214 
215 	gfp_mask = GFP_KERNEL | __GFP_ZERO;
216 	efi_pgd = (pgd_t *)__get_free_pages(gfp_mask, PGD_ALLOCATION_ORDER);
217 	if (!efi_pgd)
218 		return -ENOMEM;
219 
220 	pgd = efi_pgd + pgd_index(EFI_VA_END);
221 	p4d = p4d_alloc(&init_mm, pgd, EFI_VA_END);
222 	if (!p4d) {
223 		free_page((unsigned long)efi_pgd);
224 		return -ENOMEM;
225 	}
226 
227 	pud = pud_alloc(&init_mm, p4d, EFI_VA_END);
228 	if (!pud) {
229 		if (CONFIG_PGTABLE_LEVELS > 4)
230 			free_page((unsigned long) pgd_page_vaddr(*pgd));
231 		free_page((unsigned long)efi_pgd);
232 		return -ENOMEM;
233 	}
234 
235 	return 0;
236 }
237 
238 /*
239  * Add low kernel mappings for passing arguments to EFI functions.
240  */
241 void efi_sync_low_kernel_mappings(void)
242 {
243 	unsigned num_entries;
244 	pgd_t *pgd_k, *pgd_efi;
245 	p4d_t *p4d_k, *p4d_efi;
246 	pud_t *pud_k, *pud_efi;
247 
248 	if (efi_enabled(EFI_OLD_MEMMAP))
249 		return;
250 
251 	/*
252 	 * We can share all PGD entries apart from the one entry that
253 	 * covers the EFI runtime mapping space.
254 	 *
255 	 * Make sure the EFI runtime region mappings are guaranteed to
256 	 * only span a single PGD entry and that the entry also maps
257 	 * other important kernel regions.
258 	 */
259 	BUILD_BUG_ON(pgd_index(EFI_VA_END) != pgd_index(MODULES_END));
260 	BUILD_BUG_ON((EFI_VA_START & PGDIR_MASK) !=
261 			(EFI_VA_END & PGDIR_MASK));
262 
263 	pgd_efi = efi_pgd + pgd_index(PAGE_OFFSET);
264 	pgd_k = pgd_offset_k(PAGE_OFFSET);
265 
266 	num_entries = pgd_index(EFI_VA_END) - pgd_index(PAGE_OFFSET);
267 	memcpy(pgd_efi, pgd_k, sizeof(pgd_t) * num_entries);
268 
269 	/*
270 	 * As with PGDs, we share all P4D entries apart from the one entry
271 	 * that covers the EFI runtime mapping space.
272 	 */
273 	BUILD_BUG_ON(p4d_index(EFI_VA_END) != p4d_index(MODULES_END));
274 	BUILD_BUG_ON((EFI_VA_START & P4D_MASK) != (EFI_VA_END & P4D_MASK));
275 
276 	pgd_efi = efi_pgd + pgd_index(EFI_VA_END);
277 	pgd_k = pgd_offset_k(EFI_VA_END);
278 	p4d_efi = p4d_offset(pgd_efi, 0);
279 	p4d_k = p4d_offset(pgd_k, 0);
280 
281 	num_entries = p4d_index(EFI_VA_END);
282 	memcpy(p4d_efi, p4d_k, sizeof(p4d_t) * num_entries);
283 
284 	/*
285 	 * We share all the PUD entries apart from those that map the
286 	 * EFI regions. Copy around them.
287 	 */
288 	BUILD_BUG_ON((EFI_VA_START & ~PUD_MASK) != 0);
289 	BUILD_BUG_ON((EFI_VA_END & ~PUD_MASK) != 0);
290 
291 	p4d_efi = p4d_offset(pgd_efi, EFI_VA_END);
292 	p4d_k = p4d_offset(pgd_k, EFI_VA_END);
293 	pud_efi = pud_offset(p4d_efi, 0);
294 	pud_k = pud_offset(p4d_k, 0);
295 
296 	num_entries = pud_index(EFI_VA_END);
297 	memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
298 
299 	pud_efi = pud_offset(p4d_efi, EFI_VA_START);
300 	pud_k = pud_offset(p4d_k, EFI_VA_START);
301 
302 	num_entries = PTRS_PER_PUD - pud_index(EFI_VA_START);
303 	memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
304 }
305 
306 /*
307  * Wrapper for slow_virt_to_phys() that handles NULL addresses.
308  */
309 static inline phys_addr_t
310 virt_to_phys_or_null_size(void *va, unsigned long size)
311 {
312 	bool bad_size;
313 
314 	if (!va)
315 		return 0;
316 
317 	if (virt_addr_valid(va))
318 		return virt_to_phys(va);
319 
320 	/*
321 	 * A fully aligned variable on the stack is guaranteed not to
322 	 * cross a page bounary. Try to catch strings on the stack by
323 	 * checking that 'size' is a power of two.
324 	 */
325 	bad_size = size > PAGE_SIZE || !is_power_of_2(size);
326 
327 	WARN_ON(!IS_ALIGNED((unsigned long)va, size) || bad_size);
328 
329 	return slow_virt_to_phys(va);
330 }
331 
332 #define virt_to_phys_or_null(addr)				\
333 	virt_to_phys_or_null_size((addr), sizeof(*(addr)))
334 
335 int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)
336 {
337 	unsigned long pfn, text, pf;
338 	struct page *page;
339 	unsigned npages;
340 	pgd_t *pgd;
341 
342 	if (efi_enabled(EFI_OLD_MEMMAP))
343 		return 0;
344 
345 	/*
346 	 * Since the PGD is encrypted, set the encryption mask so that when
347 	 * this value is loaded into cr3 the PGD will be decrypted during
348 	 * the pagetable walk.
349 	 */
350 	efi_scratch.efi_pgt = (pgd_t *)__sme_pa(efi_pgd);
351 	pgd = efi_pgd;
352 
353 	/*
354 	 * It can happen that the physical address of new_memmap lands in memory
355 	 * which is not mapped in the EFI page table. Therefore we need to go
356 	 * and ident-map those pages containing the map before calling
357 	 * phys_efi_set_virtual_address_map().
358 	 */
359 	pfn = pa_memmap >> PAGE_SHIFT;
360 	pf = _PAGE_NX | _PAGE_RW | _PAGE_ENC;
361 	if (kernel_map_pages_in_pgd(pgd, pfn, pa_memmap, num_pages, pf)) {
362 		pr_err("Error ident-mapping new memmap (0x%lx)!\n", pa_memmap);
363 		return 1;
364 	}
365 
366 	efi_scratch.use_pgd = true;
367 
368 	/*
369 	 * Certain firmware versions are way too sentimential and still believe
370 	 * they are exclusive and unquestionable owners of the first physical page,
371 	 * even though they explicitly mark it as EFI_CONVENTIONAL_MEMORY
372 	 * (but then write-access it later during SetVirtualAddressMap()).
373 	 *
374 	 * Create a 1:1 mapping for this page, to avoid triple faults during early
375 	 * boot with such firmware. We are free to hand this page to the BIOS,
376 	 * as trim_bios_range() will reserve the first page and isolate it away
377 	 * from memory allocators anyway.
378 	 */
379 	pf = _PAGE_RW;
380 	if (sev_active())
381 		pf |= _PAGE_ENC;
382 
383 	if (kernel_map_pages_in_pgd(pgd, 0x0, 0x0, 1, pf)) {
384 		pr_err("Failed to create 1:1 mapping for the first page!\n");
385 		return 1;
386 	}
387 
388 	/*
389 	 * When making calls to the firmware everything needs to be 1:1
390 	 * mapped and addressable with 32-bit pointers. Map the kernel
391 	 * text and allocate a new stack because we can't rely on the
392 	 * stack pointer being < 4GB.
393 	 */
394 	if (!IS_ENABLED(CONFIG_EFI_MIXED) || efi_is_native())
395 		return 0;
396 
397 	page = alloc_page(GFP_KERNEL|__GFP_DMA32);
398 	if (!page)
399 		panic("Unable to allocate EFI runtime stack < 4GB\n");
400 
401 	efi_scratch.phys_stack = virt_to_phys(page_address(page));
402 	efi_scratch.phys_stack += PAGE_SIZE; /* stack grows down */
403 
404 	npages = (_etext - _text) >> PAGE_SHIFT;
405 	text = __pa(_text);
406 	pfn = text >> PAGE_SHIFT;
407 
408 	pf = _PAGE_RW | _PAGE_ENC;
409 	if (kernel_map_pages_in_pgd(pgd, pfn, text, npages, pf)) {
410 		pr_err("Failed to map kernel text 1:1\n");
411 		return 1;
412 	}
413 
414 	return 0;
415 }
416 
417 static void __init __map_region(efi_memory_desc_t *md, u64 va)
418 {
419 	unsigned long flags = _PAGE_RW;
420 	unsigned long pfn;
421 	pgd_t *pgd = efi_pgd;
422 
423 	if (!(md->attribute & EFI_MEMORY_WB))
424 		flags |= _PAGE_PCD;
425 
426 	if (sev_active())
427 		flags |= _PAGE_ENC;
428 
429 	pfn = md->phys_addr >> PAGE_SHIFT;
430 	if (kernel_map_pages_in_pgd(pgd, pfn, va, md->num_pages, flags))
431 		pr_warn("Error mapping PA 0x%llx -> VA 0x%llx!\n",
432 			   md->phys_addr, va);
433 }
434 
435 void __init efi_map_region(efi_memory_desc_t *md)
436 {
437 	unsigned long size = md->num_pages << PAGE_SHIFT;
438 	u64 pa = md->phys_addr;
439 
440 	if (efi_enabled(EFI_OLD_MEMMAP))
441 		return old_map_region(md);
442 
443 	/*
444 	 * Make sure the 1:1 mappings are present as a catch-all for b0rked
445 	 * firmware which doesn't update all internal pointers after switching
446 	 * to virtual mode and would otherwise crap on us.
447 	 */
448 	__map_region(md, md->phys_addr);
449 
450 	/*
451 	 * Enforce the 1:1 mapping as the default virtual address when
452 	 * booting in EFI mixed mode, because even though we may be
453 	 * running a 64-bit kernel, the firmware may only be 32-bit.
454 	 */
455 	if (!efi_is_native () && IS_ENABLED(CONFIG_EFI_MIXED)) {
456 		md->virt_addr = md->phys_addr;
457 		return;
458 	}
459 
460 	efi_va -= size;
461 
462 	/* Is PA 2M-aligned? */
463 	if (!(pa & (PMD_SIZE - 1))) {
464 		efi_va &= PMD_MASK;
465 	} else {
466 		u64 pa_offset = pa & (PMD_SIZE - 1);
467 		u64 prev_va = efi_va;
468 
469 		/* get us the same offset within this 2M page */
470 		efi_va = (efi_va & PMD_MASK) + pa_offset;
471 
472 		if (efi_va > prev_va)
473 			efi_va -= PMD_SIZE;
474 	}
475 
476 	if (efi_va < EFI_VA_END) {
477 		pr_warn(FW_WARN "VA address range overflow!\n");
478 		return;
479 	}
480 
481 	/* Do the VA map */
482 	__map_region(md, efi_va);
483 	md->virt_addr = efi_va;
484 }
485 
486 /*
487  * kexec kernel will use efi_map_region_fixed to map efi runtime memory ranges.
488  * md->virt_addr is the original virtual address which had been mapped in kexec
489  * 1st kernel.
490  */
491 void __init efi_map_region_fixed(efi_memory_desc_t *md)
492 {
493 	__map_region(md, md->phys_addr);
494 	__map_region(md, md->virt_addr);
495 }
496 
497 void __iomem *__init efi_ioremap(unsigned long phys_addr, unsigned long size,
498 				 u32 type, u64 attribute)
499 {
500 	unsigned long last_map_pfn;
501 
502 	if (type == EFI_MEMORY_MAPPED_IO)
503 		return ioremap(phys_addr, size);
504 
505 	last_map_pfn = init_memory_mapping(phys_addr, phys_addr + size);
506 	if ((last_map_pfn << PAGE_SHIFT) < phys_addr + size) {
507 		unsigned long top = last_map_pfn << PAGE_SHIFT;
508 		efi_ioremap(top, size - (top - phys_addr), type, attribute);
509 	}
510 
511 	if (!(attribute & EFI_MEMORY_WB))
512 		efi_memory_uc((u64)(unsigned long)__va(phys_addr), size);
513 
514 	return (void __iomem *)__va(phys_addr);
515 }
516 
517 void __init parse_efi_setup(u64 phys_addr, u32 data_len)
518 {
519 	efi_setup = phys_addr + sizeof(struct setup_data);
520 }
521 
522 static int __init efi_update_mappings(efi_memory_desc_t *md, unsigned long pf)
523 {
524 	unsigned long pfn;
525 	pgd_t *pgd = efi_pgd;
526 	int err1, err2;
527 
528 	/* Update the 1:1 mapping */
529 	pfn = md->phys_addr >> PAGE_SHIFT;
530 	err1 = kernel_map_pages_in_pgd(pgd, pfn, md->phys_addr, md->num_pages, pf);
531 	if (err1) {
532 		pr_err("Error while updating 1:1 mapping PA 0x%llx -> VA 0x%llx!\n",
533 			   md->phys_addr, md->virt_addr);
534 	}
535 
536 	err2 = kernel_map_pages_in_pgd(pgd, pfn, md->virt_addr, md->num_pages, pf);
537 	if (err2) {
538 		pr_err("Error while updating VA mapping PA 0x%llx -> VA 0x%llx!\n",
539 			   md->phys_addr, md->virt_addr);
540 	}
541 
542 	return err1 || err2;
543 }
544 
545 static int __init efi_update_mem_attr(struct mm_struct *mm, efi_memory_desc_t *md)
546 {
547 	unsigned long pf = 0;
548 
549 	if (md->attribute & EFI_MEMORY_XP)
550 		pf |= _PAGE_NX;
551 
552 	if (!(md->attribute & EFI_MEMORY_RO))
553 		pf |= _PAGE_RW;
554 
555 	if (sev_active())
556 		pf |= _PAGE_ENC;
557 
558 	return efi_update_mappings(md, pf);
559 }
560 
561 void __init efi_runtime_update_mappings(void)
562 {
563 	efi_memory_desc_t *md;
564 
565 	if (efi_enabled(EFI_OLD_MEMMAP)) {
566 		if (__supported_pte_mask & _PAGE_NX)
567 			runtime_code_page_mkexec();
568 		return;
569 	}
570 
571 	/*
572 	 * Use the EFI Memory Attribute Table for mapping permissions if it
573 	 * exists, since it is intended to supersede EFI_PROPERTIES_TABLE.
574 	 */
575 	if (efi_enabled(EFI_MEM_ATTR)) {
576 		efi_memattr_apply_permissions(NULL, efi_update_mem_attr);
577 		return;
578 	}
579 
580 	/*
581 	 * EFI_MEMORY_ATTRIBUTES_TABLE is intended to replace
582 	 * EFI_PROPERTIES_TABLE. So, use EFI_PROPERTIES_TABLE to update
583 	 * permissions only if EFI_MEMORY_ATTRIBUTES_TABLE is not
584 	 * published by the firmware. Even if we find a buggy implementation of
585 	 * EFI_MEMORY_ATTRIBUTES_TABLE, don't fall back to
586 	 * EFI_PROPERTIES_TABLE, because of the same reason.
587 	 */
588 
589 	if (!efi_enabled(EFI_NX_PE_DATA))
590 		return;
591 
592 	for_each_efi_memory_desc(md) {
593 		unsigned long pf = 0;
594 
595 		if (!(md->attribute & EFI_MEMORY_RUNTIME))
596 			continue;
597 
598 		if (!(md->attribute & EFI_MEMORY_WB))
599 			pf |= _PAGE_PCD;
600 
601 		if ((md->attribute & EFI_MEMORY_XP) ||
602 			(md->type == EFI_RUNTIME_SERVICES_DATA))
603 			pf |= _PAGE_NX;
604 
605 		if (!(md->attribute & EFI_MEMORY_RO) &&
606 			(md->type != EFI_RUNTIME_SERVICES_CODE))
607 			pf |= _PAGE_RW;
608 
609 		if (sev_active())
610 			pf |= _PAGE_ENC;
611 
612 		efi_update_mappings(md, pf);
613 	}
614 }
615 
616 void __init efi_dump_pagetable(void)
617 {
618 #ifdef CONFIG_EFI_PGT_DUMP
619 	if (efi_enabled(EFI_OLD_MEMMAP))
620 		ptdump_walk_pgd_level(NULL, swapper_pg_dir);
621 	else
622 		ptdump_walk_pgd_level(NULL, efi_pgd);
623 #endif
624 }
625 
626 #ifdef CONFIG_EFI_MIXED
627 extern efi_status_t efi64_thunk(u32, ...);
628 
629 #define runtime_service32(func)						 \
630 ({									 \
631 	u32 table = (u32)(unsigned long)efi.systab;			 \
632 	u32 *rt, *___f;							 \
633 									 \
634 	rt = (u32 *)(table + offsetof(efi_system_table_32_t, runtime));	 \
635 	___f = (u32 *)(*rt + offsetof(efi_runtime_services_32_t, func)); \
636 	*___f;								 \
637 })
638 
639 /*
640  * Switch to the EFI page tables early so that we can access the 1:1
641  * runtime services mappings which are not mapped in any other page
642  * tables. This function must be called before runtime_service32().
643  *
644  * Also, disable interrupts because the IDT points to 64-bit handlers,
645  * which aren't going to function correctly when we switch to 32-bit.
646  */
647 #define efi_thunk(f, ...)						\
648 ({									\
649 	efi_status_t __s;						\
650 	unsigned long __flags;						\
651 	u32 __func;							\
652 									\
653 	local_irq_save(__flags);					\
654 	arch_efi_call_virt_setup();					\
655 									\
656 	__func = runtime_service32(f);					\
657 	__s = efi64_thunk(__func, __VA_ARGS__);				\
658 									\
659 	arch_efi_call_virt_teardown();					\
660 	local_irq_restore(__flags);					\
661 									\
662 	__s;								\
663 })
664 
665 efi_status_t efi_thunk_set_virtual_address_map(
666 	void *phys_set_virtual_address_map,
667 	unsigned long memory_map_size,
668 	unsigned long descriptor_size,
669 	u32 descriptor_version,
670 	efi_memory_desc_t *virtual_map)
671 {
672 	efi_status_t status;
673 	unsigned long flags;
674 	u32 func;
675 
676 	efi_sync_low_kernel_mappings();
677 	local_irq_save(flags);
678 
679 	efi_scratch.prev_cr3 = __read_cr3();
680 	write_cr3((unsigned long)efi_scratch.efi_pgt);
681 	__flush_tlb_all();
682 
683 	func = (u32)(unsigned long)phys_set_virtual_address_map;
684 	status = efi64_thunk(func, memory_map_size, descriptor_size,
685 			     descriptor_version, virtual_map);
686 
687 	write_cr3(efi_scratch.prev_cr3);
688 	__flush_tlb_all();
689 	local_irq_restore(flags);
690 
691 	return status;
692 }
693 
694 static efi_status_t efi_thunk_get_time(efi_time_t *tm, efi_time_cap_t *tc)
695 {
696 	efi_status_t status;
697 	u32 phys_tm, phys_tc;
698 
699 	spin_lock(&rtc_lock);
700 
701 	phys_tm = virt_to_phys_or_null(tm);
702 	phys_tc = virt_to_phys_or_null(tc);
703 
704 	status = efi_thunk(get_time, phys_tm, phys_tc);
705 
706 	spin_unlock(&rtc_lock);
707 
708 	return status;
709 }
710 
711 static efi_status_t efi_thunk_set_time(efi_time_t *tm)
712 {
713 	efi_status_t status;
714 	u32 phys_tm;
715 
716 	spin_lock(&rtc_lock);
717 
718 	phys_tm = virt_to_phys_or_null(tm);
719 
720 	status = efi_thunk(set_time, phys_tm);
721 
722 	spin_unlock(&rtc_lock);
723 
724 	return status;
725 }
726 
727 static efi_status_t
728 efi_thunk_get_wakeup_time(efi_bool_t *enabled, efi_bool_t *pending,
729 			  efi_time_t *tm)
730 {
731 	efi_status_t status;
732 	u32 phys_enabled, phys_pending, phys_tm;
733 
734 	spin_lock(&rtc_lock);
735 
736 	phys_enabled = virt_to_phys_or_null(enabled);
737 	phys_pending = virt_to_phys_or_null(pending);
738 	phys_tm = virt_to_phys_or_null(tm);
739 
740 	status = efi_thunk(get_wakeup_time, phys_enabled,
741 			     phys_pending, phys_tm);
742 
743 	spin_unlock(&rtc_lock);
744 
745 	return status;
746 }
747 
748 static efi_status_t
749 efi_thunk_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm)
750 {
751 	efi_status_t status;
752 	u32 phys_tm;
753 
754 	spin_lock(&rtc_lock);
755 
756 	phys_tm = virt_to_phys_or_null(tm);
757 
758 	status = efi_thunk(set_wakeup_time, enabled, phys_tm);
759 
760 	spin_unlock(&rtc_lock);
761 
762 	return status;
763 }
764 
765 static unsigned long efi_name_size(efi_char16_t *name)
766 {
767 	return ucs2_strsize(name, EFI_VAR_NAME_LEN) + 1;
768 }
769 
770 static efi_status_t
771 efi_thunk_get_variable(efi_char16_t *name, efi_guid_t *vendor,
772 		       u32 *attr, unsigned long *data_size, void *data)
773 {
774 	efi_status_t status;
775 	u32 phys_name, phys_vendor, phys_attr;
776 	u32 phys_data_size, phys_data;
777 
778 	phys_data_size = virt_to_phys_or_null(data_size);
779 	phys_vendor = virt_to_phys_or_null(vendor);
780 	phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
781 	phys_attr = virt_to_phys_or_null(attr);
782 	phys_data = virt_to_phys_or_null_size(data, *data_size);
783 
784 	status = efi_thunk(get_variable, phys_name, phys_vendor,
785 			   phys_attr, phys_data_size, phys_data);
786 
787 	return status;
788 }
789 
790 static efi_status_t
791 efi_thunk_set_variable(efi_char16_t *name, efi_guid_t *vendor,
792 		       u32 attr, unsigned long data_size, void *data)
793 {
794 	u32 phys_name, phys_vendor, phys_data;
795 	efi_status_t status;
796 
797 	phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
798 	phys_vendor = virt_to_phys_or_null(vendor);
799 	phys_data = virt_to_phys_or_null_size(data, data_size);
800 
801 	/* If data_size is > sizeof(u32) we've got problems */
802 	status = efi_thunk(set_variable, phys_name, phys_vendor,
803 			   attr, data_size, phys_data);
804 
805 	return status;
806 }
807 
808 static efi_status_t
809 efi_thunk_get_next_variable(unsigned long *name_size,
810 			    efi_char16_t *name,
811 			    efi_guid_t *vendor)
812 {
813 	efi_status_t status;
814 	u32 phys_name_size, phys_name, phys_vendor;
815 
816 	phys_name_size = virt_to_phys_or_null(name_size);
817 	phys_vendor = virt_to_phys_or_null(vendor);
818 	phys_name = virt_to_phys_or_null_size(name, *name_size);
819 
820 	status = efi_thunk(get_next_variable, phys_name_size,
821 			   phys_name, phys_vendor);
822 
823 	return status;
824 }
825 
826 static efi_status_t
827 efi_thunk_get_next_high_mono_count(u32 *count)
828 {
829 	efi_status_t status;
830 	u32 phys_count;
831 
832 	phys_count = virt_to_phys_or_null(count);
833 	status = efi_thunk(get_next_high_mono_count, phys_count);
834 
835 	return status;
836 }
837 
838 static void
839 efi_thunk_reset_system(int reset_type, efi_status_t status,
840 		       unsigned long data_size, efi_char16_t *data)
841 {
842 	u32 phys_data;
843 
844 	phys_data = virt_to_phys_or_null_size(data, data_size);
845 
846 	efi_thunk(reset_system, reset_type, status, data_size, phys_data);
847 }
848 
849 static efi_status_t
850 efi_thunk_update_capsule(efi_capsule_header_t **capsules,
851 			 unsigned long count, unsigned long sg_list)
852 {
853 	/*
854 	 * To properly support this function we would need to repackage
855 	 * 'capsules' because the firmware doesn't understand 64-bit
856 	 * pointers.
857 	 */
858 	return EFI_UNSUPPORTED;
859 }
860 
861 static efi_status_t
862 efi_thunk_query_variable_info(u32 attr, u64 *storage_space,
863 			      u64 *remaining_space,
864 			      u64 *max_variable_size)
865 {
866 	efi_status_t status;
867 	u32 phys_storage, phys_remaining, phys_max;
868 
869 	if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
870 		return EFI_UNSUPPORTED;
871 
872 	phys_storage = virt_to_phys_or_null(storage_space);
873 	phys_remaining = virt_to_phys_or_null(remaining_space);
874 	phys_max = virt_to_phys_or_null(max_variable_size);
875 
876 	status = efi_thunk(query_variable_info, attr, phys_storage,
877 			   phys_remaining, phys_max);
878 
879 	return status;
880 }
881 
882 static efi_status_t
883 efi_thunk_query_capsule_caps(efi_capsule_header_t **capsules,
884 			     unsigned long count, u64 *max_size,
885 			     int *reset_type)
886 {
887 	/*
888 	 * To properly support this function we would need to repackage
889 	 * 'capsules' because the firmware doesn't understand 64-bit
890 	 * pointers.
891 	 */
892 	return EFI_UNSUPPORTED;
893 }
894 
895 void efi_thunk_runtime_setup(void)
896 {
897 	efi.get_time = efi_thunk_get_time;
898 	efi.set_time = efi_thunk_set_time;
899 	efi.get_wakeup_time = efi_thunk_get_wakeup_time;
900 	efi.set_wakeup_time = efi_thunk_set_wakeup_time;
901 	efi.get_variable = efi_thunk_get_variable;
902 	efi.get_next_variable = efi_thunk_get_next_variable;
903 	efi.set_variable = efi_thunk_set_variable;
904 	efi.get_next_high_mono_count = efi_thunk_get_next_high_mono_count;
905 	efi.reset_system = efi_thunk_reset_system;
906 	efi.query_variable_info = efi_thunk_query_variable_info;
907 	efi.update_capsule = efi_thunk_update_capsule;
908 	efi.query_capsule_caps = efi_thunk_query_capsule_caps;
909 }
910 #endif /* CONFIG_EFI_MIXED */
911