efi: x86: clean up previous struct mm switching
[linux-2.6-microblaze.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/memblock.h>
27 #include <linux/ioport.h>
28 #include <linux/mc146818rtc.h>
29 #include <linux/efi.h>
30 #include <linux/export.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 #include <linux/sched/task.h>
38
39 #include <asm/setup.h>
40 #include <asm/page.h>
41 #include <asm/e820/api.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 #include <asm/sev-es.h>
51
52 /*
53  * We allocate runtime services regions top-down, starting from -4G, i.e.
54  * 0xffff_ffff_0000_0000 and limit EFI VA mapping space to 64G.
55  */
56 static u64 efi_va = EFI_VA_START;
57 static struct mm_struct *efi_prev_mm;
58
59 /*
60  * We need our own copy of the higher levels of the page tables
61  * because we want to avoid inserting EFI region mappings (EFI_VA_END
62  * to EFI_VA_START) into the standard kernel page tables. Everything
63  * else can be shared, see efi_sync_low_kernel_mappings().
64  *
65  * We don't want the pgd on the pgd_list and cannot use pgd_alloc() for the
66  * allocation.
67  */
68 int __init efi_alloc_page_tables(void)
69 {
70         pgd_t *pgd, *efi_pgd;
71         p4d_t *p4d;
72         pud_t *pud;
73         gfp_t gfp_mask;
74
75         gfp_mask = GFP_KERNEL | __GFP_ZERO;
76         efi_pgd = (pgd_t *)__get_free_pages(gfp_mask, PGD_ALLOCATION_ORDER);
77         if (!efi_pgd)
78                 goto fail;
79
80         pgd = efi_pgd + pgd_index(EFI_VA_END);
81         p4d = p4d_alloc(&init_mm, pgd, EFI_VA_END);
82         if (!p4d)
83                 goto free_pgd;
84
85         pud = pud_alloc(&init_mm, p4d, EFI_VA_END);
86         if (!pud)
87                 goto free_p4d;
88
89         efi_mm.pgd = efi_pgd;
90         mm_init_cpumask(&efi_mm);
91         init_new_context(NULL, &efi_mm);
92
93         return 0;
94
95 free_p4d:
96         if (pgtable_l5_enabled())
97                 free_page((unsigned long)pgd_page_vaddr(*pgd));
98 free_pgd:
99         free_pages((unsigned long)efi_pgd, PGD_ALLOCATION_ORDER);
100 fail:
101         return -ENOMEM;
102 }
103
104 /*
105  * Add low kernel mappings for passing arguments to EFI functions.
106  */
107 void efi_sync_low_kernel_mappings(void)
108 {
109         unsigned num_entries;
110         pgd_t *pgd_k, *pgd_efi;
111         p4d_t *p4d_k, *p4d_efi;
112         pud_t *pud_k, *pud_efi;
113         pgd_t *efi_pgd = efi_mm.pgd;
114
115         /*
116          * We can share all PGD entries apart from the one entry that
117          * covers the EFI runtime mapping space.
118          *
119          * Make sure the EFI runtime region mappings are guaranteed to
120          * only span a single PGD entry and that the entry also maps
121          * other important kernel regions.
122          */
123         MAYBE_BUILD_BUG_ON(pgd_index(EFI_VA_END) != pgd_index(MODULES_END));
124         MAYBE_BUILD_BUG_ON((EFI_VA_START & PGDIR_MASK) !=
125                         (EFI_VA_END & PGDIR_MASK));
126
127         pgd_efi = efi_pgd + pgd_index(PAGE_OFFSET);
128         pgd_k = pgd_offset_k(PAGE_OFFSET);
129
130         num_entries = pgd_index(EFI_VA_END) - pgd_index(PAGE_OFFSET);
131         memcpy(pgd_efi, pgd_k, sizeof(pgd_t) * num_entries);
132
133         /*
134          * As with PGDs, we share all P4D entries apart from the one entry
135          * that covers the EFI runtime mapping space.
136          */
137         BUILD_BUG_ON(p4d_index(EFI_VA_END) != p4d_index(MODULES_END));
138         BUILD_BUG_ON((EFI_VA_START & P4D_MASK) != (EFI_VA_END & P4D_MASK));
139
140         pgd_efi = efi_pgd + pgd_index(EFI_VA_END);
141         pgd_k = pgd_offset_k(EFI_VA_END);
142         p4d_efi = p4d_offset(pgd_efi, 0);
143         p4d_k = p4d_offset(pgd_k, 0);
144
145         num_entries = p4d_index(EFI_VA_END);
146         memcpy(p4d_efi, p4d_k, sizeof(p4d_t) * num_entries);
147
148         /*
149          * We share all the PUD entries apart from those that map the
150          * EFI regions. Copy around them.
151          */
152         BUILD_BUG_ON((EFI_VA_START & ~PUD_MASK) != 0);
153         BUILD_BUG_ON((EFI_VA_END & ~PUD_MASK) != 0);
154
155         p4d_efi = p4d_offset(pgd_efi, EFI_VA_END);
156         p4d_k = p4d_offset(pgd_k, EFI_VA_END);
157         pud_efi = pud_offset(p4d_efi, 0);
158         pud_k = pud_offset(p4d_k, 0);
159
160         num_entries = pud_index(EFI_VA_END);
161         memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
162
163         pud_efi = pud_offset(p4d_efi, EFI_VA_START);
164         pud_k = pud_offset(p4d_k, EFI_VA_START);
165
166         num_entries = PTRS_PER_PUD - pud_index(EFI_VA_START);
167         memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
168 }
169
170 /*
171  * Wrapper for slow_virt_to_phys() that handles NULL addresses.
172  */
173 static inline phys_addr_t
174 virt_to_phys_or_null_size(void *va, unsigned long size)
175 {
176         phys_addr_t pa;
177
178         if (!va)
179                 return 0;
180
181         if (virt_addr_valid(va))
182                 return virt_to_phys(va);
183
184         pa = slow_virt_to_phys(va);
185
186         /* check if the object crosses a page boundary */
187         if (WARN_ON((pa ^ (pa + size - 1)) & PAGE_MASK))
188                 return 0;
189
190         return pa;
191 }
192
193 #define virt_to_phys_or_null(addr)                              \
194         virt_to_phys_or_null_size((addr), sizeof(*(addr)))
195
196 int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)
197 {
198         unsigned long pfn, text, pf, rodata;
199         struct page *page;
200         unsigned npages;
201         pgd_t *pgd = efi_mm.pgd;
202
203         /*
204          * It can happen that the physical address of new_memmap lands in memory
205          * which is not mapped in the EFI page table. Therefore we need to go
206          * and ident-map those pages containing the map before calling
207          * phys_efi_set_virtual_address_map().
208          */
209         pfn = pa_memmap >> PAGE_SHIFT;
210         pf = _PAGE_NX | _PAGE_RW | _PAGE_ENC;
211         if (kernel_map_pages_in_pgd(pgd, pfn, pa_memmap, num_pages, pf)) {
212                 pr_err("Error ident-mapping new memmap (0x%lx)!\n", pa_memmap);
213                 return 1;
214         }
215
216         /*
217          * Certain firmware versions are way too sentimential and still believe
218          * they are exclusive and unquestionable owners of the first physical page,
219          * even though they explicitly mark it as EFI_CONVENTIONAL_MEMORY
220          * (but then write-access it later during SetVirtualAddressMap()).
221          *
222          * Create a 1:1 mapping for this page, to avoid triple faults during early
223          * boot with such firmware. We are free to hand this page to the BIOS,
224          * as trim_bios_range() will reserve the first page and isolate it away
225          * from memory allocators anyway.
226          */
227         if (kernel_map_pages_in_pgd(pgd, 0x0, 0x0, 1, pf)) {
228                 pr_err("Failed to create 1:1 mapping for the first page!\n");
229                 return 1;
230         }
231
232         /*
233          * When SEV-ES is active, the GHCB as set by the kernel will be used
234          * by firmware. Create a 1:1 unencrypted mapping for each GHCB.
235          */
236         if (sev_es_efi_map_ghcbs(pgd)) {
237                 pr_err("Failed to create 1:1 mapping for the GHCBs!\n");
238                 return 1;
239         }
240
241         /*
242          * When making calls to the firmware everything needs to be 1:1
243          * mapped and addressable with 32-bit pointers. Map the kernel
244          * text and allocate a new stack because we can't rely on the
245          * stack pointer being < 4GB.
246          */
247         if (!efi_is_mixed())
248                 return 0;
249
250         page = alloc_page(GFP_KERNEL|__GFP_DMA32);
251         if (!page) {
252                 pr_err("Unable to allocate EFI runtime stack < 4GB\n");
253                 return 1;
254         }
255
256         efi_mixed_mode_stack_pa = page_to_phys(page + 1); /* stack grows down */
257
258         npages = (_etext - _text) >> PAGE_SHIFT;
259         text = __pa(_text);
260         pfn = text >> PAGE_SHIFT;
261
262         pf = _PAGE_ENC;
263         if (kernel_map_pages_in_pgd(pgd, pfn, text, npages, pf)) {
264                 pr_err("Failed to map kernel text 1:1\n");
265                 return 1;
266         }
267
268         npages = (__end_rodata - __start_rodata) >> PAGE_SHIFT;
269         rodata = __pa(__start_rodata);
270         pfn = rodata >> PAGE_SHIFT;
271
272         pf = _PAGE_NX | _PAGE_ENC;
273         if (kernel_map_pages_in_pgd(pgd, pfn, rodata, npages, pf)) {
274                 pr_err("Failed to map kernel rodata 1:1\n");
275                 return 1;
276         }
277
278         return 0;
279 }
280
281 static void __init __map_region(efi_memory_desc_t *md, u64 va)
282 {
283         unsigned long flags = _PAGE_RW;
284         unsigned long pfn;
285         pgd_t *pgd = efi_mm.pgd;
286
287         /*
288          * EFI_RUNTIME_SERVICES_CODE regions typically cover PE/COFF
289          * executable images in memory that consist of both R-X and
290          * RW- sections, so we cannot apply read-only or non-exec
291          * permissions just yet. However, modern EFI systems provide
292          * a memory attributes table that describes those sections
293          * with the appropriate restricted permissions, which are
294          * applied in efi_runtime_update_mappings() below. All other
295          * regions can be mapped non-executable at this point, with
296          * the exception of boot services code regions, but those will
297          * be unmapped again entirely in efi_free_boot_services().
298          */
299         if (md->type != EFI_BOOT_SERVICES_CODE &&
300             md->type != EFI_RUNTIME_SERVICES_CODE)
301                 flags |= _PAGE_NX;
302
303         if (!(md->attribute & EFI_MEMORY_WB))
304                 flags |= _PAGE_PCD;
305
306         if (sev_active() && md->type != EFI_MEMORY_MAPPED_IO)
307                 flags |= _PAGE_ENC;
308
309         pfn = md->phys_addr >> PAGE_SHIFT;
310         if (kernel_map_pages_in_pgd(pgd, pfn, va, md->num_pages, flags))
311                 pr_warn("Error mapping PA 0x%llx -> VA 0x%llx!\n",
312                            md->phys_addr, va);
313 }
314
315 void __init efi_map_region(efi_memory_desc_t *md)
316 {
317         unsigned long size = md->num_pages << PAGE_SHIFT;
318         u64 pa = md->phys_addr;
319
320         /*
321          * Make sure the 1:1 mappings are present as a catch-all for b0rked
322          * firmware which doesn't update all internal pointers after switching
323          * to virtual mode and would otherwise crap on us.
324          */
325         __map_region(md, md->phys_addr);
326
327         /*
328          * Enforce the 1:1 mapping as the default virtual address when
329          * booting in EFI mixed mode, because even though we may be
330          * running a 64-bit kernel, the firmware may only be 32-bit.
331          */
332         if (efi_is_mixed()) {
333                 md->virt_addr = md->phys_addr;
334                 return;
335         }
336
337         efi_va -= size;
338
339         /* Is PA 2M-aligned? */
340         if (!(pa & (PMD_SIZE - 1))) {
341                 efi_va &= PMD_MASK;
342         } else {
343                 u64 pa_offset = pa & (PMD_SIZE - 1);
344                 u64 prev_va = efi_va;
345
346                 /* get us the same offset within this 2M page */
347                 efi_va = (efi_va & PMD_MASK) + pa_offset;
348
349                 if (efi_va > prev_va)
350                         efi_va -= PMD_SIZE;
351         }
352
353         if (efi_va < EFI_VA_END) {
354                 pr_warn(FW_WARN "VA address range overflow!\n");
355                 return;
356         }
357
358         /* Do the VA map */
359         __map_region(md, efi_va);
360         md->virt_addr = efi_va;
361 }
362
363 /*
364  * kexec kernel will use efi_map_region_fixed to map efi runtime memory ranges.
365  * md->virt_addr is the original virtual address which had been mapped in kexec
366  * 1st kernel.
367  */
368 void __init efi_map_region_fixed(efi_memory_desc_t *md)
369 {
370         __map_region(md, md->phys_addr);
371         __map_region(md, md->virt_addr);
372 }
373
374 void __init parse_efi_setup(u64 phys_addr, u32 data_len)
375 {
376         efi_setup = phys_addr + sizeof(struct setup_data);
377 }
378
379 static int __init efi_update_mappings(efi_memory_desc_t *md, unsigned long pf)
380 {
381         unsigned long pfn;
382         pgd_t *pgd = efi_mm.pgd;
383         int err1, err2;
384
385         /* Update the 1:1 mapping */
386         pfn = md->phys_addr >> PAGE_SHIFT;
387         err1 = kernel_map_pages_in_pgd(pgd, pfn, md->phys_addr, md->num_pages, pf);
388         if (err1) {
389                 pr_err("Error while updating 1:1 mapping PA 0x%llx -> VA 0x%llx!\n",
390                            md->phys_addr, md->virt_addr);
391         }
392
393         err2 = kernel_map_pages_in_pgd(pgd, pfn, md->virt_addr, md->num_pages, pf);
394         if (err2) {
395                 pr_err("Error while updating VA mapping PA 0x%llx -> VA 0x%llx!\n",
396                            md->phys_addr, md->virt_addr);
397         }
398
399         return err1 || err2;
400 }
401
402 static int __init efi_update_mem_attr(struct mm_struct *mm, efi_memory_desc_t *md)
403 {
404         unsigned long pf = 0;
405
406         if (md->attribute & EFI_MEMORY_XP)
407                 pf |= _PAGE_NX;
408
409         if (!(md->attribute & EFI_MEMORY_RO))
410                 pf |= _PAGE_RW;
411
412         if (sev_active())
413                 pf |= _PAGE_ENC;
414
415         return efi_update_mappings(md, pf);
416 }
417
418 void __init efi_runtime_update_mappings(void)
419 {
420         efi_memory_desc_t *md;
421
422         /*
423          * Use the EFI Memory Attribute Table for mapping permissions if it
424          * exists, since it is intended to supersede EFI_PROPERTIES_TABLE.
425          */
426         if (efi_enabled(EFI_MEM_ATTR)) {
427                 efi_memattr_apply_permissions(NULL, efi_update_mem_attr);
428                 return;
429         }
430
431         /*
432          * EFI_MEMORY_ATTRIBUTES_TABLE is intended to replace
433          * EFI_PROPERTIES_TABLE. So, use EFI_PROPERTIES_TABLE to update
434          * permissions only if EFI_MEMORY_ATTRIBUTES_TABLE is not
435          * published by the firmware. Even if we find a buggy implementation of
436          * EFI_MEMORY_ATTRIBUTES_TABLE, don't fall back to
437          * EFI_PROPERTIES_TABLE, because of the same reason.
438          */
439
440         if (!efi_enabled(EFI_NX_PE_DATA))
441                 return;
442
443         for_each_efi_memory_desc(md) {
444                 unsigned long pf = 0;
445
446                 if (!(md->attribute & EFI_MEMORY_RUNTIME))
447                         continue;
448
449                 if (!(md->attribute & EFI_MEMORY_WB))
450                         pf |= _PAGE_PCD;
451
452                 if ((md->attribute & EFI_MEMORY_XP) ||
453                         (md->type == EFI_RUNTIME_SERVICES_DATA))
454                         pf |= _PAGE_NX;
455
456                 if (!(md->attribute & EFI_MEMORY_RO) &&
457                         (md->type != EFI_RUNTIME_SERVICES_CODE))
458                         pf |= _PAGE_RW;
459
460                 if (sev_active())
461                         pf |= _PAGE_ENC;
462
463                 efi_update_mappings(md, pf);
464         }
465 }
466
467 void __init efi_dump_pagetable(void)
468 {
469 #ifdef CONFIG_EFI_PGT_DUMP
470         ptdump_walk_pgd_level(NULL, &efi_mm);
471 #endif
472 }
473
474 /*
475  * Makes the calling thread switch to/from efi_mm context. Can be used
476  * in a kernel thread and user context. Preemption needs to remain disabled
477  * while the EFI-mm is borrowed. mmgrab()/mmdrop() is not used because the mm
478  * can not change under us.
479  * It should be ensured that there are no concurent calls to this function.
480  */
481 void efi_enter_mm(void)
482 {
483         efi_prev_mm = current->active_mm;
484         current->active_mm = &efi_mm;
485         switch_mm(efi_prev_mm, &efi_mm, NULL);
486 }
487
488 void efi_leave_mm(void)
489 {
490         current->active_mm = efi_prev_mm;
491         switch_mm(&efi_mm, efi_prev_mm, NULL);
492 }
493
494 static DEFINE_SPINLOCK(efi_runtime_lock);
495
496 /*
497  * DS and ES contain user values.  We need to save them.
498  * The 32-bit EFI code needs a valid DS, ES, and SS.  There's no
499  * need to save the old SS: __KERNEL_DS is always acceptable.
500  */
501 #define __efi_thunk(func, ...)                                          \
502 ({                                                                      \
503         unsigned short __ds, __es;                                      \
504         efi_status_t ____s;                                             \
505                                                                         \
506         savesegment(ds, __ds);                                          \
507         savesegment(es, __es);                                          \
508                                                                         \
509         loadsegment(ss, __KERNEL_DS);                                   \
510         loadsegment(ds, __KERNEL_DS);                                   \
511         loadsegment(es, __KERNEL_DS);                                   \
512                                                                         \
513         ____s = efi64_thunk(efi.runtime->mixed_mode.func, __VA_ARGS__); \
514                                                                         \
515         loadsegment(ds, __ds);                                          \
516         loadsegment(es, __es);                                          \
517                                                                         \
518         ____s ^= (____s & BIT(31)) | (____s & BIT_ULL(31)) << 32;       \
519         ____s;                                                          \
520 })
521
522 /*
523  * Switch to the EFI page tables early so that we can access the 1:1
524  * runtime services mappings which are not mapped in any other page
525  * tables.
526  *
527  * Also, disable interrupts because the IDT points to 64-bit handlers,
528  * which aren't going to function correctly when we switch to 32-bit.
529  */
530 #define efi_thunk(func...)                                              \
531 ({                                                                      \
532         efi_status_t __s;                                               \
533                                                                         \
534         arch_efi_call_virt_setup();                                     \
535                                                                         \
536         __s = __efi_thunk(func);                                        \
537                                                                         \
538         arch_efi_call_virt_teardown();                                  \
539                                                                         \
540         __s;                                                            \
541 })
542
543 static efi_status_t __init __no_sanitize_address
544 efi_thunk_set_virtual_address_map(unsigned long memory_map_size,
545                                   unsigned long descriptor_size,
546                                   u32 descriptor_version,
547                                   efi_memory_desc_t *virtual_map)
548 {
549         efi_status_t status;
550         unsigned long flags;
551
552         efi_sync_low_kernel_mappings();
553         local_irq_save(flags);
554
555         efi_enter_mm();
556
557         status = __efi_thunk(set_virtual_address_map, memory_map_size,
558                              descriptor_size, descriptor_version, virtual_map);
559
560         efi_leave_mm();
561         local_irq_restore(flags);
562
563         return status;
564 }
565
566 static efi_status_t efi_thunk_get_time(efi_time_t *tm, efi_time_cap_t *tc)
567 {
568         return EFI_UNSUPPORTED;
569 }
570
571 static efi_status_t efi_thunk_set_time(efi_time_t *tm)
572 {
573         return EFI_UNSUPPORTED;
574 }
575
576 static efi_status_t
577 efi_thunk_get_wakeup_time(efi_bool_t *enabled, efi_bool_t *pending,
578                           efi_time_t *tm)
579 {
580         return EFI_UNSUPPORTED;
581 }
582
583 static efi_status_t
584 efi_thunk_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm)
585 {
586         return EFI_UNSUPPORTED;
587 }
588
589 static unsigned long efi_name_size(efi_char16_t *name)
590 {
591         return ucs2_strsize(name, EFI_VAR_NAME_LEN) + 1;
592 }
593
594 static efi_status_t
595 efi_thunk_get_variable(efi_char16_t *name, efi_guid_t *vendor,
596                        u32 *attr, unsigned long *data_size, void *data)
597 {
598         u8 buf[24] __aligned(8);
599         efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
600         efi_status_t status;
601         u32 phys_name, phys_vendor, phys_attr;
602         u32 phys_data_size, phys_data;
603         unsigned long flags;
604
605         spin_lock_irqsave(&efi_runtime_lock, flags);
606
607         *vnd = *vendor;
608
609         phys_data_size = virt_to_phys_or_null(data_size);
610         phys_vendor = virt_to_phys_or_null(vnd);
611         phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
612         phys_attr = virt_to_phys_or_null(attr);
613         phys_data = virt_to_phys_or_null_size(data, *data_size);
614
615         if (!phys_name || (data && !phys_data))
616                 status = EFI_INVALID_PARAMETER;
617         else
618                 status = efi_thunk(get_variable, phys_name, phys_vendor,
619                                    phys_attr, phys_data_size, phys_data);
620
621         spin_unlock_irqrestore(&efi_runtime_lock, flags);
622
623         return status;
624 }
625
626 static efi_status_t
627 efi_thunk_set_variable(efi_char16_t *name, efi_guid_t *vendor,
628                        u32 attr, unsigned long data_size, void *data)
629 {
630         u8 buf[24] __aligned(8);
631         efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
632         u32 phys_name, phys_vendor, phys_data;
633         efi_status_t status;
634         unsigned long flags;
635
636         spin_lock_irqsave(&efi_runtime_lock, flags);
637
638         *vnd = *vendor;
639
640         phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
641         phys_vendor = virt_to_phys_or_null(vnd);
642         phys_data = virt_to_phys_or_null_size(data, data_size);
643
644         if (!phys_name || (data && !phys_data))
645                 status = EFI_INVALID_PARAMETER;
646         else
647                 status = efi_thunk(set_variable, phys_name, phys_vendor,
648                                    attr, data_size, phys_data);
649
650         spin_unlock_irqrestore(&efi_runtime_lock, flags);
651
652         return status;
653 }
654
655 static efi_status_t
656 efi_thunk_set_variable_nonblocking(efi_char16_t *name, efi_guid_t *vendor,
657                                    u32 attr, unsigned long data_size,
658                                    void *data)
659 {
660         u8 buf[24] __aligned(8);
661         efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
662         u32 phys_name, phys_vendor, phys_data;
663         efi_status_t status;
664         unsigned long flags;
665
666         if (!spin_trylock_irqsave(&efi_runtime_lock, flags))
667                 return EFI_NOT_READY;
668
669         *vnd = *vendor;
670
671         phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
672         phys_vendor = virt_to_phys_or_null(vnd);
673         phys_data = virt_to_phys_or_null_size(data, data_size);
674
675         if (!phys_name || (data && !phys_data))
676                 status = EFI_INVALID_PARAMETER;
677         else
678                 status = efi_thunk(set_variable, phys_name, phys_vendor,
679                                    attr, data_size, phys_data);
680
681         spin_unlock_irqrestore(&efi_runtime_lock, flags);
682
683         return status;
684 }
685
686 static efi_status_t
687 efi_thunk_get_next_variable(unsigned long *name_size,
688                             efi_char16_t *name,
689                             efi_guid_t *vendor)
690 {
691         u8 buf[24] __aligned(8);
692         efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
693         efi_status_t status;
694         u32 phys_name_size, phys_name, phys_vendor;
695         unsigned long flags;
696
697         spin_lock_irqsave(&efi_runtime_lock, flags);
698
699         *vnd = *vendor;
700
701         phys_name_size = virt_to_phys_or_null(name_size);
702         phys_vendor = virt_to_phys_or_null(vnd);
703         phys_name = virt_to_phys_or_null_size(name, *name_size);
704
705         if (!phys_name)
706                 status = EFI_INVALID_PARAMETER;
707         else
708                 status = efi_thunk(get_next_variable, phys_name_size,
709                                    phys_name, phys_vendor);
710
711         spin_unlock_irqrestore(&efi_runtime_lock, flags);
712
713         *vendor = *vnd;
714         return status;
715 }
716
717 static efi_status_t
718 efi_thunk_get_next_high_mono_count(u32 *count)
719 {
720         return EFI_UNSUPPORTED;
721 }
722
723 static void
724 efi_thunk_reset_system(int reset_type, efi_status_t status,
725                        unsigned long data_size, efi_char16_t *data)
726 {
727         u32 phys_data;
728         unsigned long flags;
729
730         spin_lock_irqsave(&efi_runtime_lock, flags);
731
732         phys_data = virt_to_phys_or_null_size(data, data_size);
733
734         efi_thunk(reset_system, reset_type, status, data_size, phys_data);
735
736         spin_unlock_irqrestore(&efi_runtime_lock, flags);
737 }
738
739 static efi_status_t
740 efi_thunk_update_capsule(efi_capsule_header_t **capsules,
741                          unsigned long count, unsigned long sg_list)
742 {
743         /*
744          * To properly support this function we would need to repackage
745          * 'capsules' because the firmware doesn't understand 64-bit
746          * pointers.
747          */
748         return EFI_UNSUPPORTED;
749 }
750
751 static efi_status_t
752 efi_thunk_query_variable_info(u32 attr, u64 *storage_space,
753                               u64 *remaining_space,
754                               u64 *max_variable_size)
755 {
756         efi_status_t status;
757         u32 phys_storage, phys_remaining, phys_max;
758         unsigned long flags;
759
760         if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
761                 return EFI_UNSUPPORTED;
762
763         spin_lock_irqsave(&efi_runtime_lock, flags);
764
765         phys_storage = virt_to_phys_or_null(storage_space);
766         phys_remaining = virt_to_phys_or_null(remaining_space);
767         phys_max = virt_to_phys_or_null(max_variable_size);
768
769         status = efi_thunk(query_variable_info, attr, phys_storage,
770                            phys_remaining, phys_max);
771
772         spin_unlock_irqrestore(&efi_runtime_lock, flags);
773
774         return status;
775 }
776
777 static efi_status_t
778 efi_thunk_query_variable_info_nonblocking(u32 attr, u64 *storage_space,
779                                           u64 *remaining_space,
780                                           u64 *max_variable_size)
781 {
782         efi_status_t status;
783         u32 phys_storage, phys_remaining, phys_max;
784         unsigned long flags;
785
786         if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
787                 return EFI_UNSUPPORTED;
788
789         if (!spin_trylock_irqsave(&efi_runtime_lock, flags))
790                 return EFI_NOT_READY;
791
792         phys_storage = virt_to_phys_or_null(storage_space);
793         phys_remaining = virt_to_phys_or_null(remaining_space);
794         phys_max = virt_to_phys_or_null(max_variable_size);
795
796         status = efi_thunk(query_variable_info, attr, phys_storage,
797                            phys_remaining, phys_max);
798
799         spin_unlock_irqrestore(&efi_runtime_lock, flags);
800
801         return status;
802 }
803
804 static efi_status_t
805 efi_thunk_query_capsule_caps(efi_capsule_header_t **capsules,
806                              unsigned long count, u64 *max_size,
807                              int *reset_type)
808 {
809         /*
810          * To properly support this function we would need to repackage
811          * 'capsules' because the firmware doesn't understand 64-bit
812          * pointers.
813          */
814         return EFI_UNSUPPORTED;
815 }
816
817 void __init efi_thunk_runtime_setup(void)
818 {
819         if (!IS_ENABLED(CONFIG_EFI_MIXED))
820                 return;
821
822         efi.get_time = efi_thunk_get_time;
823         efi.set_time = efi_thunk_set_time;
824         efi.get_wakeup_time = efi_thunk_get_wakeup_time;
825         efi.set_wakeup_time = efi_thunk_set_wakeup_time;
826         efi.get_variable = efi_thunk_get_variable;
827         efi.get_next_variable = efi_thunk_get_next_variable;
828         efi.set_variable = efi_thunk_set_variable;
829         efi.set_variable_nonblocking = efi_thunk_set_variable_nonblocking;
830         efi.get_next_high_mono_count = efi_thunk_get_next_high_mono_count;
831         efi.reset_system = efi_thunk_reset_system;
832         efi.query_variable_info = efi_thunk_query_variable_info;
833         efi.query_variable_info_nonblocking = efi_thunk_query_variable_info_nonblocking;
834         efi.update_capsule = efi_thunk_update_capsule;
835         efi.query_capsule_caps = efi_thunk_query_capsule_caps;
836 }
837
838 efi_status_t __init __no_sanitize_address
839 efi_set_virtual_address_map(unsigned long memory_map_size,
840                             unsigned long descriptor_size,
841                             u32 descriptor_version,
842                             efi_memory_desc_t *virtual_map,
843                             unsigned long systab_phys)
844 {
845         const efi_system_table_t *systab = (efi_system_table_t *)systab_phys;
846         efi_status_t status;
847         unsigned long flags;
848
849         if (efi_is_mixed())
850                 return efi_thunk_set_virtual_address_map(memory_map_size,
851                                                          descriptor_size,
852                                                          descriptor_version,
853                                                          virtual_map);
854         efi_enter_mm();
855
856         kernel_fpu_begin();
857
858         /* Disable interrupts around EFI calls: */
859         local_irq_save(flags);
860         status = efi_call(efi.runtime->set_virtual_address_map,
861                           memory_map_size, descriptor_size,
862                           descriptor_version, virtual_map);
863         local_irq_restore(flags);
864
865         kernel_fpu_end();
866
867         /* grab the virtually remapped EFI runtime services table pointer */
868         efi.runtime = READ_ONCE(systab->runtime);
869
870         efi_leave_mm();
871
872         return status;
873 }