• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
early_code_mapping_set_exec(int executable)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 
efi_call_phys_prolog(void)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 		pgd_offset_k(pgd * PGDIR_SIZE)->pgd &= ~_PAGE_NX;
138 	}
139 
140 out:
141 	__flush_tlb_all();
142 
143 	return save_pgd;
144 }
145 
efi_call_phys_epilog(pgd_t * save_pgd)146 void __init efi_call_phys_epilog(pgd_t *save_pgd)
147 {
148 	/*
149 	 * After the lock is released, the original page table is restored.
150 	 */
151 	int pgd_idx, i;
152 	int nr_pgds;
153 	pgd_t *pgd;
154 	p4d_t *p4d;
155 	pud_t *pud;
156 
157 	if (!efi_enabled(EFI_OLD_MEMMAP)) {
158 		write_cr3((unsigned long)save_pgd);
159 		__flush_tlb_all();
160 		return;
161 	}
162 
163 	nr_pgds = DIV_ROUND_UP((max_pfn << PAGE_SHIFT) , PGDIR_SIZE);
164 
165 	for (pgd_idx = 0; pgd_idx < nr_pgds; pgd_idx++) {
166 		pgd = pgd_offset_k(pgd_idx * PGDIR_SIZE);
167 		set_pgd(pgd_offset_k(pgd_idx * PGDIR_SIZE), save_pgd[pgd_idx]);
168 
169 		if (!pgd_present(*pgd))
170 			continue;
171 
172 		for (i = 0; i < PTRS_PER_P4D; i++) {
173 			p4d = p4d_offset(pgd,
174 					 pgd_idx * PGDIR_SIZE + i * P4D_SIZE);
175 
176 			if (!p4d_present(*p4d))
177 				continue;
178 
179 			pud = (pud_t *)p4d_page_vaddr(*p4d);
180 			pud_free(&init_mm, pud);
181 		}
182 
183 		p4d = (p4d_t *)pgd_page_vaddr(*pgd);
184 		p4d_free(&init_mm, p4d);
185 	}
186 
187 	kfree(save_pgd);
188 
189 	__flush_tlb_all();
190 	early_code_mapping_set_exec(0);
191 }
192 
193 static pgd_t *efi_pgd;
194 
195 /*
196  * We need our own copy of the higher levels of the page tables
197  * because we want to avoid inserting EFI region mappings (EFI_VA_END
198  * to EFI_VA_START) into the standard kernel page tables. Everything
199  * else can be shared, see efi_sync_low_kernel_mappings().
200  *
201  * We don't want the pgd on the pgd_list and cannot use pgd_alloc() for the
202  * allocation.
203  */
efi_alloc_page_tables(void)204 int __init efi_alloc_page_tables(void)
205 {
206 	pgd_t *pgd;
207 	p4d_t *p4d;
208 	pud_t *pud;
209 	gfp_t gfp_mask;
210 
211 	if (efi_enabled(EFI_OLD_MEMMAP))
212 		return 0;
213 
214 	gfp_mask = GFP_KERNEL | __GFP_ZERO;
215 	efi_pgd = (pgd_t *)__get_free_pages(gfp_mask, PGD_ALLOCATION_ORDER);
216 	if (!efi_pgd)
217 		return -ENOMEM;
218 
219 	pgd = efi_pgd + pgd_index(EFI_VA_END);
220 	p4d = p4d_alloc(&init_mm, pgd, EFI_VA_END);
221 	if (!p4d) {
222 		free_page((unsigned long)efi_pgd);
223 		return -ENOMEM;
224 	}
225 
226 	pud = pud_alloc(&init_mm, p4d, EFI_VA_END);
227 	if (!pud) {
228 		if (CONFIG_PGTABLE_LEVELS > 4)
229 			free_page((unsigned long) pgd_page_vaddr(*pgd));
230 		free_pages((unsigned long)efi_pgd, PGD_ALLOCATION_ORDER);
231 		return -ENOMEM;
232 	}
233 
234 	return 0;
235 }
236 
237 /*
238  * Add low kernel mappings for passing arguments to EFI functions.
239  */
efi_sync_low_kernel_mappings(void)240 void efi_sync_low_kernel_mappings(void)
241 {
242 	unsigned num_entries;
243 	pgd_t *pgd_k, *pgd_efi;
244 	p4d_t *p4d_k, *p4d_efi;
245 	pud_t *pud_k, *pud_efi;
246 
247 	if (efi_enabled(EFI_OLD_MEMMAP))
248 		return;
249 
250 	/*
251 	 * We can share all PGD entries apart from the one entry that
252 	 * covers the EFI runtime mapping space.
253 	 *
254 	 * Make sure the EFI runtime region mappings are guaranteed to
255 	 * only span a single PGD entry and that the entry also maps
256 	 * other important kernel regions.
257 	 */
258 	BUILD_BUG_ON(pgd_index(EFI_VA_END) != pgd_index(MODULES_END));
259 	BUILD_BUG_ON((EFI_VA_START & PGDIR_MASK) !=
260 			(EFI_VA_END & PGDIR_MASK));
261 
262 	pgd_efi = efi_pgd + pgd_index(PAGE_OFFSET);
263 	pgd_k = pgd_offset_k(PAGE_OFFSET);
264 
265 	num_entries = pgd_index(EFI_VA_END) - pgd_index(PAGE_OFFSET);
266 	memcpy(pgd_efi, pgd_k, sizeof(pgd_t) * num_entries);
267 
268 	/*
269 	 * As with PGDs, we share all P4D entries apart from the one entry
270 	 * that covers the EFI runtime mapping space.
271 	 */
272 	BUILD_BUG_ON(p4d_index(EFI_VA_END) != p4d_index(MODULES_END));
273 	BUILD_BUG_ON((EFI_VA_START & P4D_MASK) != (EFI_VA_END & P4D_MASK));
274 
275 	pgd_efi = efi_pgd + pgd_index(EFI_VA_END);
276 	pgd_k = pgd_offset_k(EFI_VA_END);
277 	p4d_efi = p4d_offset(pgd_efi, 0);
278 	p4d_k = p4d_offset(pgd_k, 0);
279 
280 	num_entries = p4d_index(EFI_VA_END);
281 	memcpy(p4d_efi, p4d_k, sizeof(p4d_t) * num_entries);
282 
283 	/*
284 	 * We share all the PUD entries apart from those that map the
285 	 * EFI regions. Copy around them.
286 	 */
287 	BUILD_BUG_ON((EFI_VA_START & ~PUD_MASK) != 0);
288 	BUILD_BUG_ON((EFI_VA_END & ~PUD_MASK) != 0);
289 
290 	p4d_efi = p4d_offset(pgd_efi, EFI_VA_END);
291 	p4d_k = p4d_offset(pgd_k, EFI_VA_END);
292 	pud_efi = pud_offset(p4d_efi, 0);
293 	pud_k = pud_offset(p4d_k, 0);
294 
295 	num_entries = pud_index(EFI_VA_END);
296 	memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
297 
298 	pud_efi = pud_offset(p4d_efi, EFI_VA_START);
299 	pud_k = pud_offset(p4d_k, EFI_VA_START);
300 
301 	num_entries = PTRS_PER_PUD - pud_index(EFI_VA_START);
302 	memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
303 }
304 
305 /*
306  * Wrapper for slow_virt_to_phys() that handles NULL addresses.
307  */
308 static inline phys_addr_t
virt_to_phys_or_null_size(void * va,unsigned long size)309 virt_to_phys_or_null_size(void *va, unsigned long size)
310 {
311 	bool bad_size;
312 
313 	if (!va)
314 		return 0;
315 
316 	if (virt_addr_valid(va))
317 		return virt_to_phys(va);
318 
319 	/*
320 	 * A fully aligned variable on the stack is guaranteed not to
321 	 * cross a page bounary. Try to catch strings on the stack by
322 	 * checking that 'size' is a power of two.
323 	 */
324 	bad_size = size > PAGE_SIZE || !is_power_of_2(size);
325 
326 	WARN_ON(!IS_ALIGNED((unsigned long)va, size) || bad_size);
327 
328 	return slow_virt_to_phys(va);
329 }
330 
331 #define virt_to_phys_or_null(addr)				\
332 	virt_to_phys_or_null_size((addr), sizeof(*(addr)))
333 
efi_setup_page_tables(unsigned long pa_memmap,unsigned num_pages)334 int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)
335 {
336 	unsigned long pfn, text, pf;
337 	struct page *page;
338 	unsigned npages;
339 	pgd_t *pgd;
340 
341 	if (efi_enabled(EFI_OLD_MEMMAP))
342 		return 0;
343 
344 	/*
345 	 * Since the PGD is encrypted, set the encryption mask so that when
346 	 * this value is loaded into cr3 the PGD will be decrypted during
347 	 * the pagetable walk.
348 	 */
349 	efi_scratch.efi_pgt = (pgd_t *)__sme_pa(efi_pgd);
350 	pgd = efi_pgd;
351 
352 	/*
353 	 * It can happen that the physical address of new_memmap lands in memory
354 	 * which is not mapped in the EFI page table. Therefore we need to go
355 	 * and ident-map those pages containing the map before calling
356 	 * phys_efi_set_virtual_address_map().
357 	 */
358 	pfn = pa_memmap >> PAGE_SHIFT;
359 	pf = _PAGE_NX | _PAGE_RW | _PAGE_ENC;
360 	if (kernel_map_pages_in_pgd(pgd, pfn, pa_memmap, num_pages, pf)) {
361 		pr_err("Error ident-mapping new memmap (0x%lx)!\n", pa_memmap);
362 		return 1;
363 	}
364 
365 	efi_scratch.use_pgd = true;
366 
367 	/*
368 	 * Certain firmware versions are way too sentimential and still believe
369 	 * they are exclusive and unquestionable owners of the first physical page,
370 	 * even though they explicitly mark it as EFI_CONVENTIONAL_MEMORY
371 	 * (but then write-access it later during SetVirtualAddressMap()).
372 	 *
373 	 * Create a 1:1 mapping for this page, to avoid triple faults during early
374 	 * boot with such firmware. We are free to hand this page to the BIOS,
375 	 * as trim_bios_range() will reserve the first page and isolate it away
376 	 * from memory allocators anyway.
377 	 */
378 	if (kernel_map_pages_in_pgd(pgd, 0x0, 0x0, 1, _PAGE_RW)) {
379 		pr_err("Failed to create 1:1 mapping for the first page!\n");
380 		return 1;
381 	}
382 
383 	/*
384 	 * When making calls to the firmware everything needs to be 1:1
385 	 * mapped and addressable with 32-bit pointers. Map the kernel
386 	 * text and allocate a new stack because we can't rely on the
387 	 * stack pointer being < 4GB.
388 	 */
389 	if (!IS_ENABLED(CONFIG_EFI_MIXED) || efi_is_native())
390 		return 0;
391 
392 	page = alloc_page(GFP_KERNEL|__GFP_DMA32);
393 	if (!page) {
394 		pr_err("Unable to allocate EFI runtime stack < 4GB\n");
395 		return 1;
396 	}
397 
398 	efi_scratch.phys_stack = page_to_phys(page + 1); /* stack grows down */
399 
400 	npages = (_etext - _text) >> PAGE_SHIFT;
401 	text = __pa(_text);
402 	pfn = text >> PAGE_SHIFT;
403 
404 	pf = _PAGE_RW | _PAGE_ENC;
405 	if (kernel_map_pages_in_pgd(pgd, pfn, text, npages, pf)) {
406 		pr_err("Failed to map kernel text 1:1\n");
407 		return 1;
408 	}
409 
410 	return 0;
411 }
412 
__map_region(efi_memory_desc_t * md,u64 va)413 static void __init __map_region(efi_memory_desc_t *md, u64 va)
414 {
415 	unsigned long flags = _PAGE_RW;
416 	unsigned long pfn;
417 	pgd_t *pgd = efi_pgd;
418 
419 	if (!(md->attribute & EFI_MEMORY_WB))
420 		flags |= _PAGE_PCD;
421 
422 	pfn = md->phys_addr >> PAGE_SHIFT;
423 	if (kernel_map_pages_in_pgd(pgd, pfn, va, md->num_pages, flags))
424 		pr_warn("Error mapping PA 0x%llx -> VA 0x%llx!\n",
425 			   md->phys_addr, va);
426 }
427 
efi_map_region(efi_memory_desc_t * md)428 void __init efi_map_region(efi_memory_desc_t *md)
429 {
430 	unsigned long size = md->num_pages << PAGE_SHIFT;
431 	u64 pa = md->phys_addr;
432 
433 	if (efi_enabled(EFI_OLD_MEMMAP))
434 		return old_map_region(md);
435 
436 	/*
437 	 * Make sure the 1:1 mappings are present as a catch-all for b0rked
438 	 * firmware which doesn't update all internal pointers after switching
439 	 * to virtual mode and would otherwise crap on us.
440 	 */
441 	__map_region(md, md->phys_addr);
442 
443 	/*
444 	 * Enforce the 1:1 mapping as the default virtual address when
445 	 * booting in EFI mixed mode, because even though we may be
446 	 * running a 64-bit kernel, the firmware may only be 32-bit.
447 	 */
448 	if (!efi_is_native () && IS_ENABLED(CONFIG_EFI_MIXED)) {
449 		md->virt_addr = md->phys_addr;
450 		return;
451 	}
452 
453 	efi_va -= size;
454 
455 	/* Is PA 2M-aligned? */
456 	if (!(pa & (PMD_SIZE - 1))) {
457 		efi_va &= PMD_MASK;
458 	} else {
459 		u64 pa_offset = pa & (PMD_SIZE - 1);
460 		u64 prev_va = efi_va;
461 
462 		/* get us the same offset within this 2M page */
463 		efi_va = (efi_va & PMD_MASK) + pa_offset;
464 
465 		if (efi_va > prev_va)
466 			efi_va -= PMD_SIZE;
467 	}
468 
469 	if (efi_va < EFI_VA_END) {
470 		pr_warn(FW_WARN "VA address range overflow!\n");
471 		return;
472 	}
473 
474 	/* Do the VA map */
475 	__map_region(md, efi_va);
476 	md->virt_addr = efi_va;
477 }
478 
479 /*
480  * kexec kernel will use efi_map_region_fixed to map efi runtime memory ranges.
481  * md->virt_addr is the original virtual address which had been mapped in kexec
482  * 1st kernel.
483  */
efi_map_region_fixed(efi_memory_desc_t * md)484 void __init efi_map_region_fixed(efi_memory_desc_t *md)
485 {
486 	__map_region(md, md->phys_addr);
487 	__map_region(md, md->virt_addr);
488 }
489 
efi_ioremap(unsigned long phys_addr,unsigned long size,u32 type,u64 attribute)490 void __iomem *__init efi_ioremap(unsigned long phys_addr, unsigned long size,
491 				 u32 type, u64 attribute)
492 {
493 	unsigned long last_map_pfn;
494 
495 	if (type == EFI_MEMORY_MAPPED_IO)
496 		return ioremap(phys_addr, size);
497 
498 	last_map_pfn = init_memory_mapping(phys_addr, phys_addr + size);
499 	if ((last_map_pfn << PAGE_SHIFT) < phys_addr + size) {
500 		unsigned long top = last_map_pfn << PAGE_SHIFT;
501 		efi_ioremap(top, size - (top - phys_addr), type, attribute);
502 	}
503 
504 	if (!(attribute & EFI_MEMORY_WB))
505 		efi_memory_uc((u64)(unsigned long)__va(phys_addr), size);
506 
507 	return (void __iomem *)__va(phys_addr);
508 }
509 
parse_efi_setup(u64 phys_addr,u32 data_len)510 void __init parse_efi_setup(u64 phys_addr, u32 data_len)
511 {
512 	efi_setup = phys_addr + sizeof(struct setup_data);
513 }
514 
efi_update_mappings(efi_memory_desc_t * md,unsigned long pf)515 static int __init efi_update_mappings(efi_memory_desc_t *md, unsigned long pf)
516 {
517 	unsigned long pfn;
518 	pgd_t *pgd = efi_pgd;
519 	int err1, err2;
520 
521 	/* Update the 1:1 mapping */
522 	pfn = md->phys_addr >> PAGE_SHIFT;
523 	err1 = kernel_map_pages_in_pgd(pgd, pfn, md->phys_addr, md->num_pages, pf);
524 	if (err1) {
525 		pr_err("Error while updating 1:1 mapping PA 0x%llx -> VA 0x%llx!\n",
526 			   md->phys_addr, md->virt_addr);
527 	}
528 
529 	err2 = kernel_map_pages_in_pgd(pgd, pfn, md->virt_addr, md->num_pages, pf);
530 	if (err2) {
531 		pr_err("Error while updating VA mapping PA 0x%llx -> VA 0x%llx!\n",
532 			   md->phys_addr, md->virt_addr);
533 	}
534 
535 	return err1 || err2;
536 }
537 
efi_update_mem_attr(struct mm_struct * mm,efi_memory_desc_t * md)538 static int __init efi_update_mem_attr(struct mm_struct *mm, efi_memory_desc_t *md)
539 {
540 	unsigned long pf = 0;
541 
542 	if (md->attribute & EFI_MEMORY_XP)
543 		pf |= _PAGE_NX;
544 
545 	if (!(md->attribute & EFI_MEMORY_RO))
546 		pf |= _PAGE_RW;
547 
548 	return efi_update_mappings(md, pf);
549 }
550 
efi_runtime_update_mappings(void)551 void __init efi_runtime_update_mappings(void)
552 {
553 	efi_memory_desc_t *md;
554 
555 	if (efi_enabled(EFI_OLD_MEMMAP)) {
556 		if (__supported_pte_mask & _PAGE_NX)
557 			runtime_code_page_mkexec();
558 		return;
559 	}
560 
561 	/*
562 	 * Use the EFI Memory Attribute Table for mapping permissions if it
563 	 * exists, since it is intended to supersede EFI_PROPERTIES_TABLE.
564 	 */
565 	if (efi_enabled(EFI_MEM_ATTR)) {
566 		efi_memattr_apply_permissions(NULL, efi_update_mem_attr);
567 		return;
568 	}
569 
570 	/*
571 	 * EFI_MEMORY_ATTRIBUTES_TABLE is intended to replace
572 	 * EFI_PROPERTIES_TABLE. So, use EFI_PROPERTIES_TABLE to update
573 	 * permissions only if EFI_MEMORY_ATTRIBUTES_TABLE is not
574 	 * published by the firmware. Even if we find a buggy implementation of
575 	 * EFI_MEMORY_ATTRIBUTES_TABLE, don't fall back to
576 	 * EFI_PROPERTIES_TABLE, because of the same reason.
577 	 */
578 
579 	if (!efi_enabled(EFI_NX_PE_DATA))
580 		return;
581 
582 	for_each_efi_memory_desc(md) {
583 		unsigned long pf = 0;
584 
585 		if (!(md->attribute & EFI_MEMORY_RUNTIME))
586 			continue;
587 
588 		if (!(md->attribute & EFI_MEMORY_WB))
589 			pf |= _PAGE_PCD;
590 
591 		if ((md->attribute & EFI_MEMORY_XP) ||
592 			(md->type == EFI_RUNTIME_SERVICES_DATA))
593 			pf |= _PAGE_NX;
594 
595 		if (!(md->attribute & EFI_MEMORY_RO) &&
596 			(md->type != EFI_RUNTIME_SERVICES_CODE))
597 			pf |= _PAGE_RW;
598 
599 		efi_update_mappings(md, pf);
600 	}
601 }
602 
efi_dump_pagetable(void)603 void __init efi_dump_pagetable(void)
604 {
605 #ifdef CONFIG_EFI_PGT_DUMP
606 	if (efi_enabled(EFI_OLD_MEMMAP))
607 		ptdump_walk_pgd_level(NULL, swapper_pg_dir);
608 	else
609 		ptdump_walk_pgd_level(NULL, efi_pgd);
610 #endif
611 }
612 
613 #ifdef CONFIG_EFI_MIXED
614 extern efi_status_t efi64_thunk(u32, ...);
615 
616 #define runtime_service32(func)						 \
617 ({									 \
618 	u32 table = (u32)(unsigned long)efi.systab;			 \
619 	u32 *rt, *___f;							 \
620 									 \
621 	rt = (u32 *)(table + offsetof(efi_system_table_32_t, runtime));	 \
622 	___f = (u32 *)(*rt + offsetof(efi_runtime_services_32_t, func)); \
623 	*___f;								 \
624 })
625 
626 /*
627  * Switch to the EFI page tables early so that we can access the 1:1
628  * runtime services mappings which are not mapped in any other page
629  * tables. This function must be called before runtime_service32().
630  *
631  * Also, disable interrupts because the IDT points to 64-bit handlers,
632  * which aren't going to function correctly when we switch to 32-bit.
633  */
634 #define efi_thunk(f, ...)						\
635 ({									\
636 	efi_status_t __s;						\
637 	unsigned long __flags;						\
638 	u32 __func;							\
639 									\
640 	local_irq_save(__flags);					\
641 	arch_efi_call_virt_setup();					\
642 									\
643 	__func = runtime_service32(f);					\
644 	__s = efi64_thunk(__func, __VA_ARGS__);				\
645 									\
646 	arch_efi_call_virt_teardown();					\
647 	local_irq_restore(__flags);					\
648 									\
649 	__s;								\
650 })
651 
efi_thunk_set_virtual_address_map(void * phys_set_virtual_address_map,unsigned long memory_map_size,unsigned long descriptor_size,u32 descriptor_version,efi_memory_desc_t * virtual_map)652 efi_status_t efi_thunk_set_virtual_address_map(
653 	void *phys_set_virtual_address_map,
654 	unsigned long memory_map_size,
655 	unsigned long descriptor_size,
656 	u32 descriptor_version,
657 	efi_memory_desc_t *virtual_map)
658 {
659 	efi_status_t status;
660 	unsigned long flags;
661 	u32 func;
662 
663 	efi_sync_low_kernel_mappings();
664 	local_irq_save(flags);
665 
666 	efi_scratch.prev_cr3 = __read_cr3();
667 	write_cr3((unsigned long)efi_scratch.efi_pgt);
668 	__flush_tlb_all();
669 
670 	func = (u32)(unsigned long)phys_set_virtual_address_map;
671 	status = efi64_thunk(func, memory_map_size, descriptor_size,
672 			     descriptor_version, virtual_map);
673 
674 	write_cr3(efi_scratch.prev_cr3);
675 	__flush_tlb_all();
676 	local_irq_restore(flags);
677 
678 	return status;
679 }
680 
efi_thunk_get_time(efi_time_t * tm,efi_time_cap_t * tc)681 static efi_status_t efi_thunk_get_time(efi_time_t *tm, efi_time_cap_t *tc)
682 {
683 	efi_status_t status;
684 	u32 phys_tm, phys_tc;
685 
686 	spin_lock(&rtc_lock);
687 
688 	phys_tm = virt_to_phys_or_null(tm);
689 	phys_tc = virt_to_phys_or_null(tc);
690 
691 	status = efi_thunk(get_time, phys_tm, phys_tc);
692 
693 	spin_unlock(&rtc_lock);
694 
695 	return status;
696 }
697 
efi_thunk_set_time(efi_time_t * tm)698 static efi_status_t efi_thunk_set_time(efi_time_t *tm)
699 {
700 	efi_status_t status;
701 	u32 phys_tm;
702 
703 	spin_lock(&rtc_lock);
704 
705 	phys_tm = virt_to_phys_or_null(tm);
706 
707 	status = efi_thunk(set_time, phys_tm);
708 
709 	spin_unlock(&rtc_lock);
710 
711 	return status;
712 }
713 
714 static efi_status_t
efi_thunk_get_wakeup_time(efi_bool_t * enabled,efi_bool_t * pending,efi_time_t * tm)715 efi_thunk_get_wakeup_time(efi_bool_t *enabled, efi_bool_t *pending,
716 			  efi_time_t *tm)
717 {
718 	efi_status_t status;
719 	u32 phys_enabled, phys_pending, phys_tm;
720 
721 	spin_lock(&rtc_lock);
722 
723 	phys_enabled = virt_to_phys_or_null(enabled);
724 	phys_pending = virt_to_phys_or_null(pending);
725 	phys_tm = virt_to_phys_or_null(tm);
726 
727 	status = efi_thunk(get_wakeup_time, phys_enabled,
728 			     phys_pending, phys_tm);
729 
730 	spin_unlock(&rtc_lock);
731 
732 	return status;
733 }
734 
735 static efi_status_t
efi_thunk_set_wakeup_time(efi_bool_t enabled,efi_time_t * tm)736 efi_thunk_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm)
737 {
738 	efi_status_t status;
739 	u32 phys_tm;
740 
741 	spin_lock(&rtc_lock);
742 
743 	phys_tm = virt_to_phys_or_null(tm);
744 
745 	status = efi_thunk(set_wakeup_time, enabled, phys_tm);
746 
747 	spin_unlock(&rtc_lock);
748 
749 	return status;
750 }
751 
efi_name_size(efi_char16_t * name)752 static unsigned long efi_name_size(efi_char16_t *name)
753 {
754 	return ucs2_strsize(name, EFI_VAR_NAME_LEN) + 1;
755 }
756 
757 static efi_status_t
efi_thunk_get_variable(efi_char16_t * name,efi_guid_t * vendor,u32 * attr,unsigned long * data_size,void * data)758 efi_thunk_get_variable(efi_char16_t *name, efi_guid_t *vendor,
759 		       u32 *attr, unsigned long *data_size, void *data)
760 {
761 	efi_status_t status;
762 	u32 phys_name, phys_vendor, phys_attr;
763 	u32 phys_data_size, phys_data;
764 
765 	phys_data_size = virt_to_phys_or_null(data_size);
766 	phys_vendor = virt_to_phys_or_null(vendor);
767 	phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
768 	phys_attr = virt_to_phys_or_null(attr);
769 	phys_data = virt_to_phys_or_null_size(data, *data_size);
770 
771 	status = efi_thunk(get_variable, phys_name, phys_vendor,
772 			   phys_attr, phys_data_size, phys_data);
773 
774 	return status;
775 }
776 
777 static efi_status_t
efi_thunk_set_variable(efi_char16_t * name,efi_guid_t * vendor,u32 attr,unsigned long data_size,void * data)778 efi_thunk_set_variable(efi_char16_t *name, efi_guid_t *vendor,
779 		       u32 attr, unsigned long data_size, void *data)
780 {
781 	u32 phys_name, phys_vendor, phys_data;
782 	efi_status_t status;
783 
784 	phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
785 	phys_vendor = virt_to_phys_or_null(vendor);
786 	phys_data = virt_to_phys_or_null_size(data, data_size);
787 
788 	/* If data_size is > sizeof(u32) we've got problems */
789 	status = efi_thunk(set_variable, phys_name, phys_vendor,
790 			   attr, data_size, phys_data);
791 
792 	return status;
793 }
794 
795 static efi_status_t
efi_thunk_get_next_variable(unsigned long * name_size,efi_char16_t * name,efi_guid_t * vendor)796 efi_thunk_get_next_variable(unsigned long *name_size,
797 			    efi_char16_t *name,
798 			    efi_guid_t *vendor)
799 {
800 	efi_status_t status;
801 	u32 phys_name_size, phys_name, phys_vendor;
802 
803 	phys_name_size = virt_to_phys_or_null(name_size);
804 	phys_vendor = virt_to_phys_or_null(vendor);
805 	phys_name = virt_to_phys_or_null_size(name, *name_size);
806 
807 	status = efi_thunk(get_next_variable, phys_name_size,
808 			   phys_name, phys_vendor);
809 
810 	return status;
811 }
812 
813 static efi_status_t
efi_thunk_get_next_high_mono_count(u32 * count)814 efi_thunk_get_next_high_mono_count(u32 *count)
815 {
816 	efi_status_t status;
817 	u32 phys_count;
818 
819 	phys_count = virt_to_phys_or_null(count);
820 	status = efi_thunk(get_next_high_mono_count, phys_count);
821 
822 	return status;
823 }
824 
825 static void
efi_thunk_reset_system(int reset_type,efi_status_t status,unsigned long data_size,efi_char16_t * data)826 efi_thunk_reset_system(int reset_type, efi_status_t status,
827 		       unsigned long data_size, efi_char16_t *data)
828 {
829 	u32 phys_data;
830 
831 	phys_data = virt_to_phys_or_null_size(data, data_size);
832 
833 	efi_thunk(reset_system, reset_type, status, data_size, phys_data);
834 }
835 
836 static efi_status_t
efi_thunk_update_capsule(efi_capsule_header_t ** capsules,unsigned long count,unsigned long sg_list)837 efi_thunk_update_capsule(efi_capsule_header_t **capsules,
838 			 unsigned long count, unsigned long sg_list)
839 {
840 	/*
841 	 * To properly support this function we would need to repackage
842 	 * 'capsules' because the firmware doesn't understand 64-bit
843 	 * pointers.
844 	 */
845 	return EFI_UNSUPPORTED;
846 }
847 
848 static efi_status_t
efi_thunk_query_variable_info(u32 attr,u64 * storage_space,u64 * remaining_space,u64 * max_variable_size)849 efi_thunk_query_variable_info(u32 attr, u64 *storage_space,
850 			      u64 *remaining_space,
851 			      u64 *max_variable_size)
852 {
853 	efi_status_t status;
854 	u32 phys_storage, phys_remaining, phys_max;
855 
856 	if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
857 		return EFI_UNSUPPORTED;
858 
859 	phys_storage = virt_to_phys_or_null(storage_space);
860 	phys_remaining = virt_to_phys_or_null(remaining_space);
861 	phys_max = virt_to_phys_or_null(max_variable_size);
862 
863 	status = efi_thunk(query_variable_info, attr, phys_storage,
864 			   phys_remaining, phys_max);
865 
866 	return status;
867 }
868 
869 static efi_status_t
efi_thunk_query_capsule_caps(efi_capsule_header_t ** capsules,unsigned long count,u64 * max_size,int * reset_type)870 efi_thunk_query_capsule_caps(efi_capsule_header_t **capsules,
871 			     unsigned long count, u64 *max_size,
872 			     int *reset_type)
873 {
874 	/*
875 	 * To properly support this function we would need to repackage
876 	 * 'capsules' because the firmware doesn't understand 64-bit
877 	 * pointers.
878 	 */
879 	return EFI_UNSUPPORTED;
880 }
881 
efi_thunk_runtime_setup(void)882 void efi_thunk_runtime_setup(void)
883 {
884 	efi.get_time = efi_thunk_get_time;
885 	efi.set_time = efi_thunk_set_time;
886 	efi.get_wakeup_time = efi_thunk_get_wakeup_time;
887 	efi.set_wakeup_time = efi_thunk_set_wakeup_time;
888 	efi.get_variable = efi_thunk_get_variable;
889 	efi.get_next_variable = efi_thunk_get_next_variable;
890 	efi.set_variable = efi_thunk_set_variable;
891 	efi.get_next_high_mono_count = efi_thunk_get_next_high_mono_count;
892 	efi.reset_system = efi_thunk_reset_system;
893 	efi.query_variable_info = efi_thunk_query_variable_info;
894 	efi.update_capsule = efi_thunk_update_capsule;
895 	efi.query_capsule_caps = efi_thunk_query_capsule_caps;
896 }
897 #endif /* CONFIG_EFI_MIXED */
898