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