Merge tag 'pm-4.15-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm
[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 int __init efi_alloc_page_tables(void)
201 {
202         pgd_t *pgd;
203         p4d_t *p4d;
204         pud_t *pud;
205         gfp_t gfp_mask;
206
207         if (efi_enabled(EFI_OLD_MEMMAP))
208                 return 0;
209
210         gfp_mask = GFP_KERNEL | __GFP_ZERO;
211         efi_pgd = (pgd_t *)__get_free_page(gfp_mask);
212         if (!efi_pgd)
213                 return -ENOMEM;
214
215         pgd = efi_pgd + pgd_index(EFI_VA_END);
216         p4d = p4d_alloc(&init_mm, pgd, EFI_VA_END);
217         if (!p4d) {
218                 free_page((unsigned long)efi_pgd);
219                 return -ENOMEM;
220         }
221
222         pud = pud_alloc(&init_mm, p4d, EFI_VA_END);
223         if (!pud) {
224                 if (CONFIG_PGTABLE_LEVELS > 4)
225                         free_page((unsigned long) pgd_page_vaddr(*pgd));
226                 free_page((unsigned long)efi_pgd);
227                 return -ENOMEM;
228         }
229
230         return 0;
231 }
232
233 /*
234  * Add low kernel mappings for passing arguments to EFI functions.
235  */
236 void efi_sync_low_kernel_mappings(void)
237 {
238         unsigned num_entries;
239         pgd_t *pgd_k, *pgd_efi;
240         p4d_t *p4d_k, *p4d_efi;
241         pud_t *pud_k, *pud_efi;
242
243         if (efi_enabled(EFI_OLD_MEMMAP))
244                 return;
245
246         /*
247          * We can share all PGD entries apart from the one entry that
248          * covers the EFI runtime mapping space.
249          *
250          * Make sure the EFI runtime region mappings are guaranteed to
251          * only span a single PGD entry and that the entry also maps
252          * other important kernel regions.
253          */
254         BUILD_BUG_ON(pgd_index(EFI_VA_END) != pgd_index(MODULES_END));
255         BUILD_BUG_ON((EFI_VA_START & PGDIR_MASK) !=
256                         (EFI_VA_END & PGDIR_MASK));
257
258         pgd_efi = efi_pgd + pgd_index(PAGE_OFFSET);
259         pgd_k = pgd_offset_k(PAGE_OFFSET);
260
261         num_entries = pgd_index(EFI_VA_END) - pgd_index(PAGE_OFFSET);
262         memcpy(pgd_efi, pgd_k, sizeof(pgd_t) * num_entries);
263
264         /*
265          * As with PGDs, we share all P4D entries apart from the one entry
266          * that covers the EFI runtime mapping space.
267          */
268         BUILD_BUG_ON(p4d_index(EFI_VA_END) != p4d_index(MODULES_END));
269         BUILD_BUG_ON((EFI_VA_START & P4D_MASK) != (EFI_VA_END & P4D_MASK));
270
271         pgd_efi = efi_pgd + pgd_index(EFI_VA_END);
272         pgd_k = pgd_offset_k(EFI_VA_END);
273         p4d_efi = p4d_offset(pgd_efi, 0);
274         p4d_k = p4d_offset(pgd_k, 0);
275
276         num_entries = p4d_index(EFI_VA_END);
277         memcpy(p4d_efi, p4d_k, sizeof(p4d_t) * num_entries);
278
279         /*
280          * We share all the PUD entries apart from those that map the
281          * EFI regions. Copy around them.
282          */
283         BUILD_BUG_ON((EFI_VA_START & ~PUD_MASK) != 0);
284         BUILD_BUG_ON((EFI_VA_END & ~PUD_MASK) != 0);
285
286         p4d_efi = p4d_offset(pgd_efi, EFI_VA_END);
287         p4d_k = p4d_offset(pgd_k, EFI_VA_END);
288         pud_efi = pud_offset(p4d_efi, 0);
289         pud_k = pud_offset(p4d_k, 0);
290
291         num_entries = pud_index(EFI_VA_END);
292         memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
293
294         pud_efi = pud_offset(p4d_efi, EFI_VA_START);
295         pud_k = pud_offset(p4d_k, EFI_VA_START);
296
297         num_entries = PTRS_PER_PUD - pud_index(EFI_VA_START);
298         memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
299 }
300
301 /*
302  * Wrapper for slow_virt_to_phys() that handles NULL addresses.
303  */
304 static inline phys_addr_t
305 virt_to_phys_or_null_size(void *va, unsigned long size)
306 {
307         bool bad_size;
308
309         if (!va)
310                 return 0;
311
312         if (virt_addr_valid(va))
313                 return virt_to_phys(va);
314
315         /*
316          * A fully aligned variable on the stack is guaranteed not to
317          * cross a page bounary. Try to catch strings on the stack by
318          * checking that 'size' is a power of two.
319          */
320         bad_size = size > PAGE_SIZE || !is_power_of_2(size);
321
322         WARN_ON(!IS_ALIGNED((unsigned long)va, size) || bad_size);
323
324         return slow_virt_to_phys(va);
325 }
326
327 #define virt_to_phys_or_null(addr)                              \
328         virt_to_phys_or_null_size((addr), sizeof(*(addr)))
329
330 int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)
331 {
332         unsigned long pfn, text, pf;
333         struct page *page;
334         unsigned npages;
335         pgd_t *pgd;
336
337         if (efi_enabled(EFI_OLD_MEMMAP))
338                 return 0;
339
340         /*
341          * Since the PGD is encrypted, set the encryption mask so that when
342          * this value is loaded into cr3 the PGD will be decrypted during
343          * the pagetable walk.
344          */
345         efi_scratch.efi_pgt = (pgd_t *)__sme_pa(efi_pgd);
346         pgd = efi_pgd;
347
348         /*
349          * It can happen that the physical address of new_memmap lands in memory
350          * which is not mapped in the EFI page table. Therefore we need to go
351          * and ident-map those pages containing the map before calling
352          * phys_efi_set_virtual_address_map().
353          */
354         pfn = pa_memmap >> PAGE_SHIFT;
355         pf = _PAGE_NX | _PAGE_RW | _PAGE_ENC;
356         if (kernel_map_pages_in_pgd(pgd, pfn, pa_memmap, num_pages, pf)) {
357                 pr_err("Error ident-mapping new memmap (0x%lx)!\n", pa_memmap);
358                 return 1;
359         }
360
361         efi_scratch.use_pgd = true;
362
363         /*
364          * Certain firmware versions are way too sentimential and still believe
365          * they are exclusive and unquestionable owners of the first physical page,
366          * even though they explicitly mark it as EFI_CONVENTIONAL_MEMORY
367          * (but then write-access it later during SetVirtualAddressMap()).
368          *
369          * Create a 1:1 mapping for this page, to avoid triple faults during early
370          * boot with such firmware. We are free to hand this page to the BIOS,
371          * as trim_bios_range() will reserve the first page and isolate it away
372          * from memory allocators anyway.
373          */
374         pf = _PAGE_RW;
375         if (sev_active())
376                 pf |= _PAGE_ENC;
377
378         if (kernel_map_pages_in_pgd(pgd, 0x0, 0x0, 1, pf)) {
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                 panic("Unable to allocate EFI runtime stack < 4GB\n");
395
396         efi_scratch.phys_stack = virt_to_phys(page_address(page));
397         efi_scratch.phys_stack += PAGE_SIZE; /* stack grows down */
398
399         npages = (_etext - _text) >> PAGE_SHIFT;
400         text = __pa(_text);
401         pfn = text >> PAGE_SHIFT;
402
403         pf = _PAGE_RW | _PAGE_ENC;
404         if (kernel_map_pages_in_pgd(pgd, pfn, text, npages, pf)) {
405                 pr_err("Failed to map kernel text 1:1\n");
406                 return 1;
407         }
408
409         return 0;
410 }
411
412 static void __init __map_region(efi_memory_desc_t *md, u64 va)
413 {
414         unsigned long flags = _PAGE_RW;
415         unsigned long pfn;
416         pgd_t *pgd = efi_pgd;
417
418         if (!(md->attribute & EFI_MEMORY_WB))
419                 flags |= _PAGE_PCD;
420
421         if (sev_active())
422                 flags |= _PAGE_ENC;
423
424         pfn = md->phys_addr >> PAGE_SHIFT;
425         if (kernel_map_pages_in_pgd(pgd, pfn, va, md->num_pages, flags))
426                 pr_warn("Error mapping PA 0x%llx -> VA 0x%llx!\n",
427                            md->phys_addr, va);
428 }
429
430 void __init efi_map_region(efi_memory_desc_t *md)
431 {
432         unsigned long size = md->num_pages << PAGE_SHIFT;
433         u64 pa = md->phys_addr;
434
435         if (efi_enabled(EFI_OLD_MEMMAP))
436                 return old_map_region(md);
437
438         /*
439          * Make sure the 1:1 mappings are present as a catch-all for b0rked
440          * firmware which doesn't update all internal pointers after switching
441          * to virtual mode and would otherwise crap on us.
442          */
443         __map_region(md, md->phys_addr);
444
445         /*
446          * Enforce the 1:1 mapping as the default virtual address when
447          * booting in EFI mixed mode, because even though we may be
448          * running a 64-bit kernel, the firmware may only be 32-bit.
449          */
450         if (!efi_is_native () && IS_ENABLED(CONFIG_EFI_MIXED)) {
451                 md->virt_addr = md->phys_addr;
452                 return;
453         }
454
455         efi_va -= size;
456
457         /* Is PA 2M-aligned? */
458         if (!(pa & (PMD_SIZE - 1))) {
459                 efi_va &= PMD_MASK;
460         } else {
461                 u64 pa_offset = pa & (PMD_SIZE - 1);
462                 u64 prev_va = efi_va;
463
464                 /* get us the same offset within this 2M page */
465                 efi_va = (efi_va & PMD_MASK) + pa_offset;
466
467                 if (efi_va > prev_va)
468                         efi_va -= PMD_SIZE;
469         }
470
471         if (efi_va < EFI_VA_END) {
472                 pr_warn(FW_WARN "VA address range overflow!\n");
473                 return;
474         }
475
476         /* Do the VA map */
477         __map_region(md, efi_va);
478         md->virt_addr = efi_va;
479 }
480
481 /*
482  * kexec kernel will use efi_map_region_fixed to map efi runtime memory ranges.
483  * md->virt_addr is the original virtual address which had been mapped in kexec
484  * 1st kernel.
485  */
486 void __init efi_map_region_fixed(efi_memory_desc_t *md)
487 {
488         __map_region(md, md->phys_addr);
489         __map_region(md, md->virt_addr);
490 }
491
492 void __iomem *__init efi_ioremap(unsigned long phys_addr, unsigned long size,
493                                  u32 type, u64 attribute)
494 {
495         unsigned long last_map_pfn;
496
497         if (type == EFI_MEMORY_MAPPED_IO)
498                 return ioremap(phys_addr, size);
499
500         last_map_pfn = init_memory_mapping(phys_addr, phys_addr + size);
501         if ((last_map_pfn << PAGE_SHIFT) < phys_addr + size) {
502                 unsigned long top = last_map_pfn << PAGE_SHIFT;
503                 efi_ioremap(top, size - (top - phys_addr), type, attribute);
504         }
505
506         if (!(attribute & EFI_MEMORY_WB))
507                 efi_memory_uc((u64)(unsigned long)__va(phys_addr), size);
508
509         return (void __iomem *)__va(phys_addr);
510 }
511
512 void __init parse_efi_setup(u64 phys_addr, u32 data_len)
513 {
514         efi_setup = phys_addr + sizeof(struct setup_data);
515 }
516
517 static int __init efi_update_mappings(efi_memory_desc_t *md, unsigned long pf)
518 {
519         unsigned long pfn;
520         pgd_t *pgd = efi_pgd;
521         int err1, err2;
522
523         /* Update the 1:1 mapping */
524         pfn = md->phys_addr >> PAGE_SHIFT;
525         err1 = kernel_map_pages_in_pgd(pgd, pfn, md->phys_addr, md->num_pages, pf);
526         if (err1) {
527                 pr_err("Error while updating 1:1 mapping PA 0x%llx -> VA 0x%llx!\n",
528                            md->phys_addr, md->virt_addr);
529         }
530
531         err2 = kernel_map_pages_in_pgd(pgd, pfn, md->virt_addr, md->num_pages, pf);
532         if (err2) {
533                 pr_err("Error while updating VA mapping PA 0x%llx -> VA 0x%llx!\n",
534                            md->phys_addr, md->virt_addr);
535         }
536
537         return err1 || err2;
538 }
539
540 static int __init efi_update_mem_attr(struct mm_struct *mm, efi_memory_desc_t *md)
541 {
542         unsigned long pf = 0;
543
544         if (md->attribute & EFI_MEMORY_XP)
545                 pf |= _PAGE_NX;
546
547         if (!(md->attribute & EFI_MEMORY_RO))
548                 pf |= _PAGE_RW;
549
550         if (sev_active())
551                 pf |= _PAGE_ENC;
552
553         return efi_update_mappings(md, pf);
554 }
555
556 void __init efi_runtime_update_mappings(void)
557 {
558         efi_memory_desc_t *md;
559
560         if (efi_enabled(EFI_OLD_MEMMAP)) {
561                 if (__supported_pte_mask & _PAGE_NX)
562                         runtime_code_page_mkexec();
563                 return;
564         }
565
566         /*
567          * Use the EFI Memory Attribute Table for mapping permissions if it
568          * exists, since it is intended to supersede EFI_PROPERTIES_TABLE.
569          */
570         if (efi_enabled(EFI_MEM_ATTR)) {
571                 efi_memattr_apply_permissions(NULL, efi_update_mem_attr);
572                 return;
573         }
574
575         /*
576          * EFI_MEMORY_ATTRIBUTES_TABLE is intended to replace
577          * EFI_PROPERTIES_TABLE. So, use EFI_PROPERTIES_TABLE to update
578          * permissions only if EFI_MEMORY_ATTRIBUTES_TABLE is not
579          * published by the firmware. Even if we find a buggy implementation of
580          * EFI_MEMORY_ATTRIBUTES_TABLE, don't fall back to
581          * EFI_PROPERTIES_TABLE, because of the same reason.
582          */
583
584         if (!efi_enabled(EFI_NX_PE_DATA))
585                 return;
586
587         for_each_efi_memory_desc(md) {
588                 unsigned long pf = 0;
589
590                 if (!(md->attribute & EFI_MEMORY_RUNTIME))
591                         continue;
592
593                 if (!(md->attribute & EFI_MEMORY_WB))
594                         pf |= _PAGE_PCD;
595
596                 if ((md->attribute & EFI_MEMORY_XP) ||
597                         (md->type == EFI_RUNTIME_SERVICES_DATA))
598                         pf |= _PAGE_NX;
599
600                 if (!(md->attribute & EFI_MEMORY_RO) &&
601                         (md->type != EFI_RUNTIME_SERVICES_CODE))
602                         pf |= _PAGE_RW;
603
604                 if (sev_active())
605                         pf |= _PAGE_ENC;
606
607                 efi_update_mappings(md, pf);
608         }
609 }
610
611 void __init efi_dump_pagetable(void)
612 {
613 #ifdef CONFIG_EFI_PGT_DUMP
614         if (efi_enabled(EFI_OLD_MEMMAP))
615                 ptdump_walk_pgd_level(NULL, swapper_pg_dir);
616         else
617                 ptdump_walk_pgd_level(NULL, efi_pgd);
618 #endif
619 }
620
621 #ifdef CONFIG_EFI_MIXED
622 extern efi_status_t efi64_thunk(u32, ...);
623
624 #define runtime_service32(func)                                          \
625 ({                                                                       \
626         u32 table = (u32)(unsigned long)efi.systab;                      \
627         u32 *rt, *___f;                                                  \
628                                                                          \
629         rt = (u32 *)(table + offsetof(efi_system_table_32_t, runtime));  \
630         ___f = (u32 *)(*rt + offsetof(efi_runtime_services_32_t, func)); \
631         *___f;                                                           \
632 })
633
634 /*
635  * Switch to the EFI page tables early so that we can access the 1:1
636  * runtime services mappings which are not mapped in any other page
637  * tables. This function must be called before runtime_service32().
638  *
639  * Also, disable interrupts because the IDT points to 64-bit handlers,
640  * which aren't going to function correctly when we switch to 32-bit.
641  */
642 #define efi_thunk(f, ...)                                               \
643 ({                                                                      \
644         efi_status_t __s;                                               \
645         unsigned long __flags;                                          \
646         u32 __func;                                                     \
647                                                                         \
648         local_irq_save(__flags);                                        \
649         arch_efi_call_virt_setup();                                     \
650                                                                         \
651         __func = runtime_service32(f);                                  \
652         __s = efi64_thunk(__func, __VA_ARGS__);                         \
653                                                                         \
654         arch_efi_call_virt_teardown();                                  \
655         local_irq_restore(__flags);                                     \
656                                                                         \
657         __s;                                                            \
658 })
659
660 efi_status_t efi_thunk_set_virtual_address_map(
661         void *phys_set_virtual_address_map,
662         unsigned long memory_map_size,
663         unsigned long descriptor_size,
664         u32 descriptor_version,
665         efi_memory_desc_t *virtual_map)
666 {
667         efi_status_t status;
668         unsigned long flags;
669         u32 func;
670
671         efi_sync_low_kernel_mappings();
672         local_irq_save(flags);
673
674         efi_scratch.prev_cr3 = __read_cr3();
675         write_cr3((unsigned long)efi_scratch.efi_pgt);
676         __flush_tlb_all();
677
678         func = (u32)(unsigned long)phys_set_virtual_address_map;
679         status = efi64_thunk(func, memory_map_size, descriptor_size,
680                              descriptor_version, virtual_map);
681
682         write_cr3(efi_scratch.prev_cr3);
683         __flush_tlb_all();
684         local_irq_restore(flags);
685
686         return status;
687 }
688
689 static efi_status_t efi_thunk_get_time(efi_time_t *tm, efi_time_cap_t *tc)
690 {
691         efi_status_t status;
692         u32 phys_tm, phys_tc;
693
694         spin_lock(&rtc_lock);
695
696         phys_tm = virt_to_phys_or_null(tm);
697         phys_tc = virt_to_phys_or_null(tc);
698
699         status = efi_thunk(get_time, phys_tm, phys_tc);
700
701         spin_unlock(&rtc_lock);
702
703         return status;
704 }
705
706 static efi_status_t efi_thunk_set_time(efi_time_t *tm)
707 {
708         efi_status_t status;
709         u32 phys_tm;
710
711         spin_lock(&rtc_lock);
712
713         phys_tm = virt_to_phys_or_null(tm);
714
715         status = efi_thunk(set_time, phys_tm);
716
717         spin_unlock(&rtc_lock);
718
719         return status;
720 }
721
722 static efi_status_t
723 efi_thunk_get_wakeup_time(efi_bool_t *enabled, efi_bool_t *pending,
724                           efi_time_t *tm)
725 {
726         efi_status_t status;
727         u32 phys_enabled, phys_pending, phys_tm;
728
729         spin_lock(&rtc_lock);
730
731         phys_enabled = virt_to_phys_or_null(enabled);
732         phys_pending = virt_to_phys_or_null(pending);
733         phys_tm = virt_to_phys_or_null(tm);
734
735         status = efi_thunk(get_wakeup_time, phys_enabled,
736                              phys_pending, phys_tm);
737
738         spin_unlock(&rtc_lock);
739
740         return status;
741 }
742
743 static efi_status_t
744 efi_thunk_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm)
745 {
746         efi_status_t status;
747         u32 phys_tm;
748
749         spin_lock(&rtc_lock);
750
751         phys_tm = virt_to_phys_or_null(tm);
752
753         status = efi_thunk(set_wakeup_time, enabled, phys_tm);
754
755         spin_unlock(&rtc_lock);
756
757         return status;
758 }
759
760 static unsigned long efi_name_size(efi_char16_t *name)
761 {
762         return ucs2_strsize(name, EFI_VAR_NAME_LEN) + 1;
763 }
764
765 static efi_status_t
766 efi_thunk_get_variable(efi_char16_t *name, efi_guid_t *vendor,
767                        u32 *attr, unsigned long *data_size, void *data)
768 {
769         efi_status_t status;
770         u32 phys_name, phys_vendor, phys_attr;
771         u32 phys_data_size, phys_data;
772
773         phys_data_size = virt_to_phys_or_null(data_size);
774         phys_vendor = virt_to_phys_or_null(vendor);
775         phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
776         phys_attr = virt_to_phys_or_null(attr);
777         phys_data = virt_to_phys_or_null_size(data, *data_size);
778
779         status = efi_thunk(get_variable, phys_name, phys_vendor,
780                            phys_attr, phys_data_size, phys_data);
781
782         return status;
783 }
784
785 static efi_status_t
786 efi_thunk_set_variable(efi_char16_t *name, efi_guid_t *vendor,
787                        u32 attr, unsigned long data_size, void *data)
788 {
789         u32 phys_name, phys_vendor, phys_data;
790         efi_status_t status;
791
792         phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
793         phys_vendor = virt_to_phys_or_null(vendor);
794         phys_data = virt_to_phys_or_null_size(data, data_size);
795
796         /* If data_size is > sizeof(u32) we've got problems */
797         status = efi_thunk(set_variable, phys_name, phys_vendor,
798                            attr, data_size, phys_data);
799
800         return status;
801 }
802
803 static efi_status_t
804 efi_thunk_get_next_variable(unsigned long *name_size,
805                             efi_char16_t *name,
806                             efi_guid_t *vendor)
807 {
808         efi_status_t status;
809         u32 phys_name_size, phys_name, phys_vendor;
810
811         phys_name_size = virt_to_phys_or_null(name_size);
812         phys_vendor = virt_to_phys_or_null(vendor);
813         phys_name = virt_to_phys_or_null_size(name, *name_size);
814
815         status = efi_thunk(get_next_variable, phys_name_size,
816                            phys_name, phys_vendor);
817
818         return status;
819 }
820
821 static efi_status_t
822 efi_thunk_get_next_high_mono_count(u32 *count)
823 {
824         efi_status_t status;
825         u32 phys_count;
826
827         phys_count = virt_to_phys_or_null(count);
828         status = efi_thunk(get_next_high_mono_count, phys_count);
829
830         return status;
831 }
832
833 static void
834 efi_thunk_reset_system(int reset_type, efi_status_t status,
835                        unsigned long data_size, efi_char16_t *data)
836 {
837         u32 phys_data;
838
839         phys_data = virt_to_phys_or_null_size(data, data_size);
840
841         efi_thunk(reset_system, reset_type, status, data_size, phys_data);
842 }
843
844 static efi_status_t
845 efi_thunk_update_capsule(efi_capsule_header_t **capsules,
846                          unsigned long count, unsigned long sg_list)
847 {
848         /*
849          * To properly support this function we would need to repackage
850          * 'capsules' because the firmware doesn't understand 64-bit
851          * pointers.
852          */
853         return EFI_UNSUPPORTED;
854 }
855
856 static efi_status_t
857 efi_thunk_query_variable_info(u32 attr, u64 *storage_space,
858                               u64 *remaining_space,
859                               u64 *max_variable_size)
860 {
861         efi_status_t status;
862         u32 phys_storage, phys_remaining, phys_max;
863
864         if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
865                 return EFI_UNSUPPORTED;
866
867         phys_storage = virt_to_phys_or_null(storage_space);
868         phys_remaining = virt_to_phys_or_null(remaining_space);
869         phys_max = virt_to_phys_or_null(max_variable_size);
870
871         status = efi_thunk(query_variable_info, attr, phys_storage,
872                            phys_remaining, phys_max);
873
874         return status;
875 }
876
877 static efi_status_t
878 efi_thunk_query_capsule_caps(efi_capsule_header_t **capsules,
879                              unsigned long count, u64 *max_size,
880                              int *reset_type)
881 {
882         /*
883          * To properly support this function we would need to repackage
884          * 'capsules' because the firmware doesn't understand 64-bit
885          * pointers.
886          */
887         return EFI_UNSUPPORTED;
888 }
889
890 void efi_thunk_runtime_setup(void)
891 {
892         efi.get_time = efi_thunk_get_time;
893         efi.set_time = efi_thunk_set_time;
894         efi.get_wakeup_time = efi_thunk_get_wakeup_time;
895         efi.set_wakeup_time = efi_thunk_set_wakeup_time;
896         efi.get_variable = efi_thunk_get_variable;
897         efi.get_next_variable = efi_thunk_get_next_variable;
898         efi.set_variable = efi_thunk_set_variable;
899         efi.get_next_high_mono_count = efi_thunk_get_next_high_mono_count;
900         efi.reset_system = efi_thunk_reset_system;
901         efi.query_variable_info = efi_thunk_query_variable_info;
902         efi.update_capsule = efi_thunk_update_capsule;
903         efi.query_capsule_caps = efi_thunk_query_capsule_caps;
904 }
905 #endif /* CONFIG_EFI_MIXED */