Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
5b83683f HY |
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 | ||
26d7f65f MF |
19 | #define pr_fmt(fmt) "efi: " fmt |
20 | ||
5b83683f HY |
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> | |
57c8a661 | 26 | #include <linux/memblock.h> |
5b83683f | 27 | #include <linux/ioport.h> |
5ab788d7 | 28 | #include <linux/mc146818rtc.h> |
5b83683f | 29 | #include <linux/efi.h> |
116fef64 | 30 | #include <linux/export.h> |
5b83683f HY |
31 | #include <linux/uaccess.h> |
32 | #include <linux/io.h> | |
33 | #include <linux/reboot.h> | |
0d01ff25 | 34 | #include <linux/slab.h> |
f6697df3 | 35 | #include <linux/ucs2_string.h> |
4d96f910 | 36 | #include <linux/cc_platform.h> |
03781e40 | 37 | #include <linux/sched/task.h> |
5b83683f HY |
38 | |
39 | #include <asm/setup.h> | |
40 | #include <asm/page.h> | |
66441bd3 | 41 | #include <asm/e820/api.h> |
5b83683f | 42 | #include <asm/tlbflush.h> |
5b83683f HY |
43 | #include <asm/proto.h> |
44 | #include <asm/efi.h> | |
4de0d4a6 | 45 | #include <asm/cacheflush.h> |
3819cd48 | 46 | #include <asm/fixmap.h> |
d2f7cbe7 | 47 | #include <asm/realmode.h> |
4f9dbcfc | 48 | #include <asm/time.h> |
67a9108e | 49 | #include <asm/pgalloc.h> |
e759959f | 50 | #include <asm/sev.h> |
5b83683f | 51 | |
d2f7cbe7 | 52 | /* |
b1d17761 | 53 | * We allocate runtime services regions top-down, starting from -4G, i.e. |
d2f7cbe7 BP |
54 | * 0xffff_ffff_0000_0000 and limit EFI VA mapping space to 64G. |
55 | */ | |
8266e31e | 56 | static u64 efi_va = EFI_VA_START; |
514b1a84 | 57 | static struct mm_struct *efi_prev_mm; |
67a9108e MF |
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(). | |
d9e9a641 DH |
64 | * |
65 | * We don't want the pgd on the pgd_list and cannot use pgd_alloc() for the | |
66 | * allocation. | |
67a9108e MF |
67 | */ |
68 | int __init efi_alloc_page_tables(void) | |
69 | { | |
3ede3417 | 70 | pgd_t *pgd, *efi_pgd; |
e981316f | 71 | p4d_t *p4d; |
67a9108e MF |
72 | pud_t *pud; |
73 | gfp_t gfp_mask; | |
74 | ||
75f296d9 | 75 | gfp_mask = GFP_KERNEL | __GFP_ZERO; |
d9e9a641 | 76 | efi_pgd = (pgd_t *)__get_free_pages(gfp_mask, PGD_ALLOCATION_ORDER); |
67a9108e | 77 | if (!efi_pgd) |
c2fe61d8 | 78 | goto fail; |
67a9108e MF |
79 | |
80 | pgd = efi_pgd + pgd_index(EFI_VA_END); | |
e981316f | 81 | p4d = p4d_alloc(&init_mm, pgd, EFI_VA_END); |
c2fe61d8 AS |
82 | if (!p4d) |
83 | goto free_pgd; | |
67a9108e | 84 | |
e981316f | 85 | pud = pud_alloc(&init_mm, p4d, EFI_VA_END); |
c2fe61d8 AS |
86 | if (!pud) |
87 | goto free_p4d; | |
67a9108e | 88 | |
3ede3417 | 89 | efi_mm.pgd = efi_pgd; |
7e904a91 SP |
90 | mm_init_cpumask(&efi_mm); |
91 | init_new_context(NULL, &efi_mm); | |
92 | ||
67a9108e | 93 | return 0; |
c2fe61d8 AS |
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; | |
67a9108e MF |
102 | } |
103 | ||
d2f7cbe7 BP |
104 | /* |
105 | * Add low kernel mappings for passing arguments to EFI functions. | |
106 | */ | |
107 | void efi_sync_low_kernel_mappings(void) | |
108 | { | |
67a9108e MF |
109 | unsigned num_entries; |
110 | pgd_t *pgd_k, *pgd_efi; | |
e0c4f675 | 111 | p4d_t *p4d_k, *p4d_efi; |
67a9108e | 112 | pud_t *pud_k, *pud_efi; |
3ede3417 | 113 | pgd_t *efi_pgd = efi_mm.pgd; |
d2f7cbe7 | 114 | |
67a9108e MF |
115 | pgd_efi = efi_pgd + pgd_index(PAGE_OFFSET); |
116 | pgd_k = pgd_offset_k(PAGE_OFFSET); | |
117 | ||
118 | num_entries = pgd_index(EFI_VA_END) - pgd_index(PAGE_OFFSET); | |
119 | memcpy(pgd_efi, pgd_k, sizeof(pgd_t) * num_entries); | |
d2f7cbe7 | 120 | |
e981316f KS |
121 | pgd_efi = efi_pgd + pgd_index(EFI_VA_END); |
122 | pgd_k = pgd_offset_k(EFI_VA_END); | |
123 | p4d_efi = p4d_offset(pgd_efi, 0); | |
124 | p4d_k = p4d_offset(pgd_k, 0); | |
125 | ||
126 | num_entries = p4d_index(EFI_VA_END); | |
127 | memcpy(p4d_efi, p4d_k, sizeof(p4d_t) * num_entries); | |
128 | ||
67a9108e MF |
129 | /* |
130 | * We share all the PUD entries apart from those that map the | |
131 | * EFI regions. Copy around them. | |
132 | */ | |
133 | BUILD_BUG_ON((EFI_VA_START & ~PUD_MASK) != 0); | |
134 | BUILD_BUG_ON((EFI_VA_END & ~PUD_MASK) != 0); | |
135 | ||
e981316f KS |
136 | p4d_efi = p4d_offset(pgd_efi, EFI_VA_END); |
137 | p4d_k = p4d_offset(pgd_k, EFI_VA_END); | |
e0c4f675 | 138 | pud_efi = pud_offset(p4d_efi, 0); |
e0c4f675 | 139 | pud_k = pud_offset(p4d_k, 0); |
67a9108e MF |
140 | |
141 | num_entries = pud_index(EFI_VA_END); | |
142 | memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries); | |
143 | ||
e0c4f675 | 144 | pud_efi = pud_offset(p4d_efi, EFI_VA_START); |
e0c4f675 | 145 | pud_k = pud_offset(p4d_k, EFI_VA_START); |
67a9108e MF |
146 | |
147 | num_entries = PTRS_PER_PUD - pud_index(EFI_VA_START); | |
148 | memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries); | |
d2f7cbe7 BP |
149 | } |
150 | ||
f6697df3 MF |
151 | /* |
152 | * Wrapper for slow_virt_to_phys() that handles NULL addresses. | |
153 | */ | |
154 | static inline phys_addr_t | |
155 | virt_to_phys_or_null_size(void *va, unsigned long size) | |
156 | { | |
8319e9d5 | 157 | phys_addr_t pa; |
f6697df3 MF |
158 | |
159 | if (!va) | |
160 | return 0; | |
161 | ||
162 | if (virt_addr_valid(va)) | |
163 | return virt_to_phys(va); | |
164 | ||
8319e9d5 | 165 | pa = slow_virt_to_phys(va); |
f6697df3 | 166 | |
8319e9d5 AB |
167 | /* check if the object crosses a page boundary */ |
168 | if (WARN_ON((pa ^ (pa + size - 1)) & PAGE_MASK)) | |
169 | return 0; | |
f6697df3 | 170 | |
8319e9d5 | 171 | return pa; |
f6697df3 MF |
172 | } |
173 | ||
174 | #define virt_to_phys_or_null(addr) \ | |
175 | virt_to_phys_or_null_size((addr), sizeof(*(addr))) | |
176 | ||
4e78eb05 | 177 | int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages) |
d2f7cbe7 | 178 | { |
6c3a9c9a AB |
179 | extern const u8 __efi64_thunk_ret_tramp[]; |
180 | unsigned long pfn, text, pf, rodata, tramp; | |
4f9dbcfc | 181 | struct page *page; |
994448f1 | 182 | unsigned npages; |
3ede3417 | 183 | pgd_t *pgd = efi_mm.pgd; |
b7b898ae | 184 | |
b7b898ae BP |
185 | /* |
186 | * It can happen that the physical address of new_memmap lands in memory | |
187 | * which is not mapped in the EFI page table. Therefore we need to go | |
188 | * and ident-map those pages containing the map before calling | |
189 | * phys_efi_set_virtual_address_map(). | |
190 | */ | |
edc3b912 | 191 | pfn = pa_memmap >> PAGE_SHIFT; |
38eecccd TL |
192 | pf = _PAGE_NX | _PAGE_RW | _PAGE_ENC; |
193 | if (kernel_map_pages_in_pgd(pgd, pfn, pa_memmap, num_pages, pf)) { | |
b7b898ae BP |
194 | pr_err("Error ident-mapping new memmap (0x%lx)!\n", pa_memmap); |
195 | return 1; | |
196 | } | |
197 | ||
bf29bddf | 198 | /* |
d9f6e12f | 199 | * Certain firmware versions are way too sentimental and still believe |
bf29bddf JK |
200 | * they are exclusive and unquestionable owners of the first physical page, |
201 | * even though they explicitly mark it as EFI_CONVENTIONAL_MEMORY | |
202 | * (but then write-access it later during SetVirtualAddressMap()). | |
203 | * | |
204 | * Create a 1:1 mapping for this page, to avoid triple faults during early | |
205 | * boot with such firmware. We are free to hand this page to the BIOS, | |
206 | * as trim_bios_range() will reserve the first page and isolate it away | |
207 | * from memory allocators anyway. | |
208 | */ | |
1379edd5 | 209 | if (kernel_map_pages_in_pgd(pgd, 0x0, 0x0, 1, pf)) { |
bf29bddf JK |
210 | pr_err("Failed to create 1:1 mapping for the first page!\n"); |
211 | return 1; | |
212 | } | |
213 | ||
39336f4f TL |
214 | /* |
215 | * When SEV-ES is active, the GHCB as set by the kernel will be used | |
216 | * by firmware. Create a 1:1 unencrypted mapping for each GHCB. | |
217 | */ | |
218 | if (sev_es_efi_map_ghcbs(pgd)) { | |
219 | pr_err("Failed to create 1:1 mapping for the GHCBs!\n"); | |
220 | return 1; | |
221 | } | |
222 | ||
4f9dbcfc MF |
223 | /* |
224 | * When making calls to the firmware everything needs to be 1:1 | |
225 | * mapped and addressable with 32-bit pointers. Map the kernel | |
226 | * text and allocate a new stack because we can't rely on the | |
227 | * stack pointer being < 4GB. | |
228 | */ | |
a8147dba | 229 | if (!efi_is_mixed()) |
994448f1 | 230 | return 0; |
4f9dbcfc MF |
231 | |
232 | page = alloc_page(GFP_KERNEL|__GFP_DMA32); | |
e2d68a95 AB |
233 | if (!page) { |
234 | pr_err("Unable to allocate EFI runtime stack < 4GB\n"); | |
235 | return 1; | |
236 | } | |
4f9dbcfc | 237 | |
3e1e00c0 | 238 | efi_mixed_mode_stack_pa = page_to_phys(page + 1); /* stack grows down */ |
4f9dbcfc | 239 | |
f6103162 | 240 | npages = (_etext - _text) >> PAGE_SHIFT; |
4f9dbcfc MF |
241 | text = __pa(_text); |
242 | ||
6c3a9c9a AB |
243 | if (kernel_unmap_pages_in_pgd(pgd, text, npages)) { |
244 | pr_err("Failed to unmap kernel text 1:1 mapping\n"); | |
994448f1 | 245 | return 1; |
4f9dbcfc | 246 | } |
b7b898ae | 247 | |
f6103162 AB |
248 | npages = (__end_rodata - __start_rodata) >> PAGE_SHIFT; |
249 | rodata = __pa(__start_rodata); | |
250 | pfn = rodata >> PAGE_SHIFT; | |
c8502eb2 AS |
251 | |
252 | pf = _PAGE_NX | _PAGE_ENC; | |
f6103162 AB |
253 | if (kernel_map_pages_in_pgd(pgd, pfn, rodata, npages, pf)) { |
254 | pr_err("Failed to map kernel rodata 1:1\n"); | |
255 | return 1; | |
256 | } | |
257 | ||
6c3a9c9a AB |
258 | tramp = __pa(__efi64_thunk_ret_tramp); |
259 | pfn = tramp >> PAGE_SHIFT; | |
260 | ||
261 | pf = _PAGE_ENC; | |
262 | if (kernel_map_pages_in_pgd(pgd, pfn, tramp, 1, pf)) { | |
263 | pr_err("Failed to map mixed mode return trampoline\n"); | |
264 | return 1; | |
265 | } | |
266 | ||
b7b898ae BP |
267 | return 0; |
268 | } | |
269 | ||
d2f7cbe7 BP |
270 | static void __init __map_region(efi_memory_desc_t *md, u64 va) |
271 | { | |
15f003d2 | 272 | unsigned long flags = _PAGE_RW; |
edc3b912 | 273 | unsigned long pfn; |
3ede3417 | 274 | pgd_t *pgd = efi_mm.pgd; |
d2f7cbe7 | 275 | |
97bb9cdc AB |
276 | /* |
277 | * EFI_RUNTIME_SERVICES_CODE regions typically cover PE/COFF | |
278 | * executable images in memory that consist of both R-X and | |
279 | * RW- sections, so we cannot apply read-only or non-exec | |
280 | * permissions just yet. However, modern EFI systems provide | |
281 | * a memory attributes table that describes those sections | |
282 | * with the appropriate restricted permissions, which are | |
283 | * applied in efi_runtime_update_mappings() below. All other | |
284 | * regions can be mapped non-executable at this point, with | |
285 | * the exception of boot services code regions, but those will | |
286 | * be unmapped again entirely in efi_free_boot_services(). | |
287 | */ | |
288 | if (md->type != EFI_BOOT_SERVICES_CODE && | |
289 | md->type != EFI_RUNTIME_SERVICES_CODE) | |
290 | flags |= _PAGE_NX; | |
291 | ||
d2f7cbe7 | 292 | if (!(md->attribute & EFI_MEMORY_WB)) |
edc3b912 | 293 | flags |= _PAGE_PCD; |
d2f7cbe7 | 294 | |
4d96f910 TL |
295 | if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT) && |
296 | md->type != EFI_MEMORY_MAPPED_IO) | |
1379edd5 TL |
297 | flags |= _PAGE_ENC; |
298 | ||
edc3b912 MF |
299 | pfn = md->phys_addr >> PAGE_SHIFT; |
300 | if (kernel_map_pages_in_pgd(pgd, pfn, va, md->num_pages, flags)) | |
d2f7cbe7 BP |
301 | pr_warn("Error mapping PA 0x%llx -> VA 0x%llx!\n", |
302 | md->phys_addr, va); | |
303 | } | |
304 | ||
305 | void __init efi_map_region(efi_memory_desc_t *md) | |
306 | { | |
307 | unsigned long size = md->num_pages << PAGE_SHIFT; | |
308 | u64 pa = md->phys_addr; | |
309 | ||
d2f7cbe7 BP |
310 | /* |
311 | * Make sure the 1:1 mappings are present as a catch-all for b0rked | |
312 | * firmware which doesn't update all internal pointers after switching | |
313 | * to virtual mode and would otherwise crap on us. | |
314 | */ | |
315 | __map_region(md, md->phys_addr); | |
316 | ||
4f9dbcfc MF |
317 | /* |
318 | * Enforce the 1:1 mapping as the default virtual address when | |
319 | * booting in EFI mixed mode, because even though we may be | |
320 | * running a 64-bit kernel, the firmware may only be 32-bit. | |
321 | */ | |
a8147dba | 322 | if (efi_is_mixed()) { |
4f9dbcfc MF |
323 | md->virt_addr = md->phys_addr; |
324 | return; | |
325 | } | |
326 | ||
d2f7cbe7 BP |
327 | efi_va -= size; |
328 | ||
329 | /* Is PA 2M-aligned? */ | |
330 | if (!(pa & (PMD_SIZE - 1))) { | |
331 | efi_va &= PMD_MASK; | |
332 | } else { | |
333 | u64 pa_offset = pa & (PMD_SIZE - 1); | |
334 | u64 prev_va = efi_va; | |
335 | ||
336 | /* get us the same offset within this 2M page */ | |
337 | efi_va = (efi_va & PMD_MASK) + pa_offset; | |
338 | ||
339 | if (efi_va > prev_va) | |
340 | efi_va -= PMD_SIZE; | |
341 | } | |
342 | ||
343 | if (efi_va < EFI_VA_END) { | |
344 | pr_warn(FW_WARN "VA address range overflow!\n"); | |
345 | return; | |
346 | } | |
347 | ||
348 | /* Do the VA map */ | |
349 | __map_region(md, efi_va); | |
350 | md->virt_addr = efi_va; | |
351 | } | |
352 | ||
3b266496 DY |
353 | /* |
354 | * kexec kernel will use efi_map_region_fixed to map efi runtime memory ranges. | |
355 | * md->virt_addr is the original virtual address which had been mapped in kexec | |
356 | * 1st kernel. | |
357 | */ | |
358 | void __init efi_map_region_fixed(efi_memory_desc_t *md) | |
359 | { | |
0513fe1d | 360 | __map_region(md, md->phys_addr); |
3b266496 DY |
361 | __map_region(md, md->virt_addr); |
362 | } | |
363 | ||
1fec0533 DY |
364 | void __init parse_efi_setup(u64 phys_addr, u32 data_len) |
365 | { | |
366 | efi_setup = phys_addr + sizeof(struct setup_data); | |
1fec0533 | 367 | } |
c55d016f | 368 | |
18141e89 | 369 | static int __init efi_update_mappings(efi_memory_desc_t *md, unsigned long pf) |
c55d016f | 370 | { |
6d0cc887 | 371 | unsigned long pfn; |
3ede3417 | 372 | pgd_t *pgd = efi_mm.pgd; |
18141e89 SP |
373 | int err1, err2; |
374 | ||
375 | /* Update the 1:1 mapping */ | |
376 | pfn = md->phys_addr >> PAGE_SHIFT; | |
377 | err1 = kernel_map_pages_in_pgd(pgd, pfn, md->phys_addr, md->num_pages, pf); | |
378 | if (err1) { | |
379 | pr_err("Error while updating 1:1 mapping PA 0x%llx -> VA 0x%llx!\n", | |
380 | md->phys_addr, md->virt_addr); | |
381 | } | |
382 | ||
383 | err2 = kernel_map_pages_in_pgd(pgd, pfn, md->virt_addr, md->num_pages, pf); | |
384 | if (err2) { | |
385 | pr_err("Error while updating VA mapping PA 0x%llx -> VA 0x%llx!\n", | |
386 | md->phys_addr, md->virt_addr); | |
387 | } | |
388 | ||
389 | return err1 || err2; | |
390 | } | |
391 | ||
93be2859 AB |
392 | bool efi_disable_ibt_for_runtime __ro_after_init = true; |
393 | ||
cf1d2ffc AB |
394 | static int __init efi_update_mem_attr(struct mm_struct *mm, efi_memory_desc_t *md, |
395 | bool has_ibt) | |
18141e89 SP |
396 | { |
397 | unsigned long pf = 0; | |
398 | ||
93be2859 AB |
399 | efi_disable_ibt_for_runtime |= !has_ibt; |
400 | ||
18141e89 SP |
401 | if (md->attribute & EFI_MEMORY_XP) |
402 | pf |= _PAGE_NX; | |
403 | ||
404 | if (!(md->attribute & EFI_MEMORY_RO)) | |
405 | pf |= _PAGE_RW; | |
406 | ||
4d96f910 | 407 | if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT)) |
1379edd5 TL |
408 | pf |= _PAGE_ENC; |
409 | ||
18141e89 SP |
410 | return efi_update_mappings(md, pf); |
411 | } | |
412 | ||
413 | void __init efi_runtime_update_mappings(void) | |
414 | { | |
6d0cc887 | 415 | efi_memory_desc_t *md; |
6d0cc887 | 416 | |
18141e89 SP |
417 | /* |
418 | * Use the EFI Memory Attribute Table for mapping permissions if it | |
419 | * exists, since it is intended to supersede EFI_PROPERTIES_TABLE. | |
420 | */ | |
421 | if (efi_enabled(EFI_MEM_ATTR)) { | |
93be2859 | 422 | efi_disable_ibt_for_runtime = false; |
18141e89 SP |
423 | efi_memattr_apply_permissions(NULL, efi_update_mem_attr); |
424 | return; | |
425 | } | |
426 | ||
427 | /* | |
428 | * EFI_MEMORY_ATTRIBUTES_TABLE is intended to replace | |
429 | * EFI_PROPERTIES_TABLE. So, use EFI_PROPERTIES_TABLE to update | |
430 | * permissions only if EFI_MEMORY_ATTRIBUTES_TABLE is not | |
431 | * published by the firmware. Even if we find a buggy implementation of | |
432 | * EFI_MEMORY_ATTRIBUTES_TABLE, don't fall back to | |
433 | * EFI_PROPERTIES_TABLE, because of the same reason. | |
434 | */ | |
435 | ||
6d0cc887 | 436 | if (!efi_enabled(EFI_NX_PE_DATA)) |
c55d016f BP |
437 | return; |
438 | ||
78ce248f | 439 | for_each_efi_memory_desc(md) { |
6d0cc887 | 440 | unsigned long pf = 0; |
6d0cc887 SP |
441 | |
442 | if (!(md->attribute & EFI_MEMORY_RUNTIME)) | |
443 | continue; | |
444 | ||
445 | if (!(md->attribute & EFI_MEMORY_WB)) | |
446 | pf |= _PAGE_PCD; | |
447 | ||
448 | if ((md->attribute & EFI_MEMORY_XP) || | |
449 | (md->type == EFI_RUNTIME_SERVICES_DATA)) | |
450 | pf |= _PAGE_NX; | |
451 | ||
452 | if (!(md->attribute & EFI_MEMORY_RO) && | |
453 | (md->type != EFI_RUNTIME_SERVICES_CODE)) | |
454 | pf |= _PAGE_RW; | |
455 | ||
4d96f910 | 456 | if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT)) |
1379edd5 TL |
457 | pf |= _PAGE_ENC; |
458 | ||
18141e89 | 459 | efi_update_mappings(md, pf); |
6d0cc887 | 460 | } |
c55d016f | 461 | } |
11cc8512 BP |
462 | |
463 | void __init efi_dump_pagetable(void) | |
464 | { | |
465 | #ifdef CONFIG_EFI_PGT_DUMP | |
66d67fec | 466 | ptdump_walk_pgd_level(NULL, &efi_mm); |
11cc8512 BP |
467 | #endif |
468 | } | |
994448f1 | 469 | |
03781e40 SP |
470 | /* |
471 | * Makes the calling thread switch to/from efi_mm context. Can be used | |
4eda1117 SAS |
472 | * in a kernel thread and user context. Preemption needs to remain disabled |
473 | * while the EFI-mm is borrowed. mmgrab()/mmdrop() is not used because the mm | |
474 | * can not change under us. | |
d9f6e12f | 475 | * It should be ensured that there are no concurrent calls to this function. |
03781e40 | 476 | */ |
762f169f | 477 | static void efi_enter_mm(void) |
514b1a84 AB |
478 | { |
479 | efi_prev_mm = current->active_mm; | |
480 | current->active_mm = &efi_mm; | |
481 | switch_mm(efi_prev_mm, &efi_mm, NULL); | |
482 | } | |
483 | ||
762f169f | 484 | static void efi_leave_mm(void) |
03781e40 | 485 | { |
514b1a84 AB |
486 | current->active_mm = efi_prev_mm; |
487 | switch_mm(&efi_mm, efi_prev_mm, NULL); | |
03781e40 SP |
488 | } |
489 | ||
762f169f AB |
490 | void arch_efi_call_virt_setup(void) |
491 | { | |
492 | efi_sync_low_kernel_mappings(); | |
493 | efi_fpu_begin(); | |
494 | firmware_restrict_branch_speculation_start(); | |
495 | efi_enter_mm(); | |
496 | } | |
497 | ||
498 | void arch_efi_call_virt_teardown(void) | |
499 | { | |
500 | efi_leave_mm(); | |
501 | firmware_restrict_branch_speculation_end(); | |
502 | efi_fpu_end(); | |
503 | } | |
504 | ||
83a0a2ea AB |
505 | static DEFINE_SPINLOCK(efi_runtime_lock); |
506 | ||
ea5e1919 AB |
507 | /* |
508 | * DS and ES contain user values. We need to save them. | |
509 | * The 32-bit EFI code needs a valid DS, ES, and SS. There's no | |
510 | * need to save the old SS: __KERNEL_DS is always acceptable. | |
511 | */ | |
512 | #define __efi_thunk(func, ...) \ | |
513 | ({ \ | |
ea5e1919 AB |
514 | unsigned short __ds, __es; \ |
515 | efi_status_t ____s; \ | |
516 | \ | |
ea5e1919 AB |
517 | savesegment(ds, __ds); \ |
518 | savesegment(es, __es); \ | |
519 | \ | |
520 | loadsegment(ss, __KERNEL_DS); \ | |
521 | loadsegment(ds, __KERNEL_DS); \ | |
522 | loadsegment(es, __KERNEL_DS); \ | |
523 | \ | |
59f2a619 | 524 | ____s = efi64_thunk(efi.runtime->mixed_mode.func, __VA_ARGS__); \ |
ea5e1919 AB |
525 | \ |
526 | loadsegment(ds, __ds); \ | |
527 | loadsegment(es, __es); \ | |
528 | \ | |
529 | ____s ^= (____s & BIT(31)) | (____s & BIT_ULL(31)) << 32; \ | |
530 | ____s; \ | |
4f9dbcfc MF |
531 | }) |
532 | ||
533 | /* | |
534 | * Switch to the EFI page tables early so that we can access the 1:1 | |
535 | * runtime services mappings which are not mapped in any other page | |
ea5e1919 | 536 | * tables. |
4f9dbcfc MF |
537 | * |
538 | * Also, disable interrupts because the IDT points to 64-bit handlers, | |
539 | * which aren't going to function correctly when we switch to 32-bit. | |
540 | */ | |
ea5e1919 | 541 | #define efi_thunk(func...) \ |
4f9dbcfc MF |
542 | ({ \ |
543 | efi_status_t __s; \ | |
4f9dbcfc | 544 | \ |
21f86625 | 545 | arch_efi_call_virt_setup(); \ |
4f9dbcfc | 546 | \ |
ea5e1919 | 547 | __s = __efi_thunk(func); \ |
4f9dbcfc | 548 | \ |
21f86625 | 549 | arch_efi_call_virt_teardown(); \ |
4f9dbcfc MF |
550 | \ |
551 | __s; \ | |
552 | }) | |
553 | ||
3cc02861 | 554 | static efi_status_t __init __no_sanitize_address |
ea5e1919 AB |
555 | efi_thunk_set_virtual_address_map(unsigned long memory_map_size, |
556 | unsigned long descriptor_size, | |
557 | u32 descriptor_version, | |
558 | efi_memory_desc_t *virtual_map) | |
4f9dbcfc MF |
559 | { |
560 | efi_status_t status; | |
561 | unsigned long flags; | |
4f9dbcfc MF |
562 | |
563 | efi_sync_low_kernel_mappings(); | |
564 | local_irq_save(flags); | |
565 | ||
514b1a84 | 566 | efi_enter_mm(); |
4f9dbcfc | 567 | |
ea5e1919 AB |
568 | status = __efi_thunk(set_virtual_address_map, memory_map_size, |
569 | descriptor_size, descriptor_version, virtual_map); | |
4f9dbcfc | 570 | |
514b1a84 | 571 | efi_leave_mm(); |
4f9dbcfc MF |
572 | local_irq_restore(flags); |
573 | ||
574 | return status; | |
575 | } | |
576 | ||
577 | static efi_status_t efi_thunk_get_time(efi_time_t *tm, efi_time_cap_t *tc) | |
578 | { | |
f80c9f64 | 579 | return EFI_UNSUPPORTED; |
4f9dbcfc MF |
580 | } |
581 | ||
582 | static efi_status_t efi_thunk_set_time(efi_time_t *tm) | |
583 | { | |
f80c9f64 | 584 | return EFI_UNSUPPORTED; |
4f9dbcfc MF |
585 | } |
586 | ||
587 | static efi_status_t | |
588 | efi_thunk_get_wakeup_time(efi_bool_t *enabled, efi_bool_t *pending, | |
589 | efi_time_t *tm) | |
590 | { | |
f80c9f64 | 591 | return EFI_UNSUPPORTED; |
4f9dbcfc MF |
592 | } |
593 | ||
594 | static efi_status_t | |
595 | efi_thunk_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm) | |
596 | { | |
f80c9f64 | 597 | return EFI_UNSUPPORTED; |
4f9dbcfc MF |
598 | } |
599 | ||
f6697df3 MF |
600 | static unsigned long efi_name_size(efi_char16_t *name) |
601 | { | |
602 | return ucs2_strsize(name, EFI_VAR_NAME_LEN) + 1; | |
603 | } | |
4f9dbcfc MF |
604 | |
605 | static efi_status_t | |
606 | efi_thunk_get_variable(efi_char16_t *name, efi_guid_t *vendor, | |
607 | u32 *attr, unsigned long *data_size, void *data) | |
608 | { | |
63056e8b AB |
609 | u8 buf[24] __aligned(8); |
610 | efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd)); | |
4f9dbcfc MF |
611 | efi_status_t status; |
612 | u32 phys_name, phys_vendor, phys_attr; | |
613 | u32 phys_data_size, phys_data; | |
83a0a2ea AB |
614 | unsigned long flags; |
615 | ||
616 | spin_lock_irqsave(&efi_runtime_lock, flags); | |
4f9dbcfc | 617 | |
63056e8b AB |
618 | *vnd = *vendor; |
619 | ||
f6697df3 | 620 | phys_data_size = virt_to_phys_or_null(data_size); |
63056e8b | 621 | phys_vendor = virt_to_phys_or_null(vnd); |
f6697df3 MF |
622 | phys_name = virt_to_phys_or_null_size(name, efi_name_size(name)); |
623 | phys_attr = virt_to_phys_or_null(attr); | |
624 | phys_data = virt_to_phys_or_null_size(data, *data_size); | |
4f9dbcfc | 625 | |
8319e9d5 AB |
626 | if (!phys_name || (data && !phys_data)) |
627 | status = EFI_INVALID_PARAMETER; | |
628 | else | |
629 | status = efi_thunk(get_variable, phys_name, phys_vendor, | |
630 | phys_attr, phys_data_size, phys_data); | |
4f9dbcfc | 631 | |
83a0a2ea AB |
632 | spin_unlock_irqrestore(&efi_runtime_lock, flags); |
633 | ||
4f9dbcfc MF |
634 | return status; |
635 | } | |
636 | ||
637 | static efi_status_t | |
638 | efi_thunk_set_variable(efi_char16_t *name, efi_guid_t *vendor, | |
639 | u32 attr, unsigned long data_size, void *data) | |
640 | { | |
63056e8b AB |
641 | u8 buf[24] __aligned(8); |
642 | efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd)); | |
4f9dbcfc MF |
643 | u32 phys_name, phys_vendor, phys_data; |
644 | efi_status_t status; | |
83a0a2ea AB |
645 | unsigned long flags; |
646 | ||
647 | spin_lock_irqsave(&efi_runtime_lock, flags); | |
648 | ||
63056e8b AB |
649 | *vnd = *vendor; |
650 | ||
83a0a2ea | 651 | phys_name = virt_to_phys_or_null_size(name, efi_name_size(name)); |
63056e8b | 652 | phys_vendor = virt_to_phys_or_null(vnd); |
83a0a2ea AB |
653 | phys_data = virt_to_phys_or_null_size(data, data_size); |
654 | ||
a4b81ccf | 655 | if (!phys_name || (data && !phys_data)) |
8319e9d5 AB |
656 | status = EFI_INVALID_PARAMETER; |
657 | else | |
658 | status = efi_thunk(set_variable, phys_name, phys_vendor, | |
659 | attr, data_size, phys_data); | |
83a0a2ea AB |
660 | |
661 | spin_unlock_irqrestore(&efi_runtime_lock, flags); | |
662 | ||
663 | return status; | |
664 | } | |
665 | ||
666 | static efi_status_t | |
667 | efi_thunk_set_variable_nonblocking(efi_char16_t *name, efi_guid_t *vendor, | |
668 | u32 attr, unsigned long data_size, | |
669 | void *data) | |
670 | { | |
63056e8b AB |
671 | u8 buf[24] __aligned(8); |
672 | efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd)); | |
83a0a2ea AB |
673 | u32 phys_name, phys_vendor, phys_data; |
674 | efi_status_t status; | |
675 | unsigned long flags; | |
676 | ||
677 | if (!spin_trylock_irqsave(&efi_runtime_lock, flags)) | |
678 | return EFI_NOT_READY; | |
4f9dbcfc | 679 | |
63056e8b AB |
680 | *vnd = *vendor; |
681 | ||
f6697df3 | 682 | phys_name = virt_to_phys_or_null_size(name, efi_name_size(name)); |
63056e8b | 683 | phys_vendor = virt_to_phys_or_null(vnd); |
f6697df3 | 684 | phys_data = virt_to_phys_or_null_size(data, data_size); |
4f9dbcfc | 685 | |
a4b81ccf | 686 | if (!phys_name || (data && !phys_data)) |
8319e9d5 AB |
687 | status = EFI_INVALID_PARAMETER; |
688 | else | |
689 | status = efi_thunk(set_variable, phys_name, phys_vendor, | |
690 | attr, data_size, phys_data); | |
4f9dbcfc | 691 | |
83a0a2ea AB |
692 | spin_unlock_irqrestore(&efi_runtime_lock, flags); |
693 | ||
4f9dbcfc MF |
694 | return status; |
695 | } | |
696 | ||
697 | static efi_status_t | |
698 | efi_thunk_get_next_variable(unsigned long *name_size, | |
699 | efi_char16_t *name, | |
700 | efi_guid_t *vendor) | |
701 | { | |
63056e8b AB |
702 | u8 buf[24] __aligned(8); |
703 | efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd)); | |
4f9dbcfc MF |
704 | efi_status_t status; |
705 | u32 phys_name_size, phys_name, phys_vendor; | |
83a0a2ea AB |
706 | unsigned long flags; |
707 | ||
708 | spin_lock_irqsave(&efi_runtime_lock, flags); | |
4f9dbcfc | 709 | |
63056e8b AB |
710 | *vnd = *vendor; |
711 | ||
f6697df3 | 712 | phys_name_size = virt_to_phys_or_null(name_size); |
63056e8b | 713 | phys_vendor = virt_to_phys_or_null(vnd); |
f6697df3 | 714 | phys_name = virt_to_phys_or_null_size(name, *name_size); |
4f9dbcfc | 715 | |
8319e9d5 AB |
716 | if (!phys_name) |
717 | status = EFI_INVALID_PARAMETER; | |
718 | else | |
719 | status = efi_thunk(get_next_variable, phys_name_size, | |
720 | phys_name, phys_vendor); | |
4f9dbcfc | 721 | |
83a0a2ea AB |
722 | spin_unlock_irqrestore(&efi_runtime_lock, flags); |
723 | ||
63056e8b | 724 | *vendor = *vnd; |
4f9dbcfc MF |
725 | return status; |
726 | } | |
727 | ||
728 | static efi_status_t | |
729 | efi_thunk_get_next_high_mono_count(u32 *count) | |
730 | { | |
f80c9f64 | 731 | return EFI_UNSUPPORTED; |
4f9dbcfc MF |
732 | } |
733 | ||
734 | static void | |
735 | efi_thunk_reset_system(int reset_type, efi_status_t status, | |
736 | unsigned long data_size, efi_char16_t *data) | |
737 | { | |
738 | u32 phys_data; | |
83a0a2ea AB |
739 | unsigned long flags; |
740 | ||
741 | spin_lock_irqsave(&efi_runtime_lock, flags); | |
4f9dbcfc | 742 | |
f6697df3 | 743 | phys_data = virt_to_phys_or_null_size(data, data_size); |
4f9dbcfc MF |
744 | |
745 | efi_thunk(reset_system, reset_type, status, data_size, phys_data); | |
83a0a2ea AB |
746 | |
747 | spin_unlock_irqrestore(&efi_runtime_lock, flags); | |
4f9dbcfc MF |
748 | } |
749 | ||
750 | static efi_status_t | |
751 | efi_thunk_update_capsule(efi_capsule_header_t **capsules, | |
752 | unsigned long count, unsigned long sg_list) | |
753 | { | |
754 | /* | |
755 | * To properly support this function we would need to repackage | |
756 | * 'capsules' because the firmware doesn't understand 64-bit | |
757 | * pointers. | |
758 | */ | |
759 | return EFI_UNSUPPORTED; | |
760 | } | |
761 | ||
762 | static efi_status_t | |
763 | efi_thunk_query_variable_info(u32 attr, u64 *storage_space, | |
764 | u64 *remaining_space, | |
765 | u64 *max_variable_size) | |
766 | { | |
767 | efi_status_t status; | |
768 | u32 phys_storage, phys_remaining, phys_max; | |
83a0a2ea | 769 | unsigned long flags; |
4f9dbcfc MF |
770 | |
771 | if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION) | |
772 | return EFI_UNSUPPORTED; | |
773 | ||
83a0a2ea AB |
774 | spin_lock_irqsave(&efi_runtime_lock, flags); |
775 | ||
f6697df3 MF |
776 | phys_storage = virt_to_phys_or_null(storage_space); |
777 | phys_remaining = virt_to_phys_or_null(remaining_space); | |
778 | phys_max = virt_to_phys_or_null(max_variable_size); | |
4f9dbcfc | 779 | |
9a11040f | 780 | status = efi_thunk(query_variable_info, attr, phys_storage, |
4f9dbcfc MF |
781 | phys_remaining, phys_max); |
782 | ||
83a0a2ea AB |
783 | spin_unlock_irqrestore(&efi_runtime_lock, flags); |
784 | ||
785 | return status; | |
786 | } | |
787 | ||
788 | static efi_status_t | |
789 | efi_thunk_query_variable_info_nonblocking(u32 attr, u64 *storage_space, | |
790 | u64 *remaining_space, | |
791 | u64 *max_variable_size) | |
792 | { | |
793 | efi_status_t status; | |
794 | u32 phys_storage, phys_remaining, phys_max; | |
795 | unsigned long flags; | |
796 | ||
797 | if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION) | |
798 | return EFI_UNSUPPORTED; | |
799 | ||
800 | if (!spin_trylock_irqsave(&efi_runtime_lock, flags)) | |
801 | return EFI_NOT_READY; | |
802 | ||
803 | phys_storage = virt_to_phys_or_null(storage_space); | |
804 | phys_remaining = virt_to_phys_or_null(remaining_space); | |
805 | phys_max = virt_to_phys_or_null(max_variable_size); | |
806 | ||
807 | status = efi_thunk(query_variable_info, attr, phys_storage, | |
808 | phys_remaining, phys_max); | |
809 | ||
810 | spin_unlock_irqrestore(&efi_runtime_lock, flags); | |
811 | ||
4f9dbcfc MF |
812 | return status; |
813 | } | |
814 | ||
815 | static efi_status_t | |
816 | efi_thunk_query_capsule_caps(efi_capsule_header_t **capsules, | |
817 | unsigned long count, u64 *max_size, | |
818 | int *reset_type) | |
819 | { | |
820 | /* | |
821 | * To properly support this function we would need to repackage | |
822 | * 'capsules' because the firmware doesn't understand 64-bit | |
823 | * pointers. | |
824 | */ | |
825 | return EFI_UNSUPPORTED; | |
826 | } | |
827 | ||
ea5e1919 | 828 | void __init efi_thunk_runtime_setup(void) |
4f9dbcfc | 829 | { |
ea5e1919 AB |
830 | if (!IS_ENABLED(CONFIG_EFI_MIXED)) |
831 | return; | |
832 | ||
4f9dbcfc MF |
833 | efi.get_time = efi_thunk_get_time; |
834 | efi.set_time = efi_thunk_set_time; | |
835 | efi.get_wakeup_time = efi_thunk_get_wakeup_time; | |
836 | efi.set_wakeup_time = efi_thunk_set_wakeup_time; | |
837 | efi.get_variable = efi_thunk_get_variable; | |
838 | efi.get_next_variable = efi_thunk_get_next_variable; | |
839 | efi.set_variable = efi_thunk_set_variable; | |
83a0a2ea | 840 | efi.set_variable_nonblocking = efi_thunk_set_variable_nonblocking; |
4f9dbcfc MF |
841 | efi.get_next_high_mono_count = efi_thunk_get_next_high_mono_count; |
842 | efi.reset_system = efi_thunk_reset_system; | |
843 | efi.query_variable_info = efi_thunk_query_variable_info; | |
83a0a2ea | 844 | efi.query_variable_info_nonblocking = efi_thunk_query_variable_info_nonblocking; |
4f9dbcfc MF |
845 | efi.update_capsule = efi_thunk_update_capsule; |
846 | efi.query_capsule_caps = efi_thunk_query_capsule_caps; | |
847 | } | |
69829470 | 848 | |
3cc02861 AB |
849 | efi_status_t __init __no_sanitize_address |
850 | efi_set_virtual_address_map(unsigned long memory_map_size, | |
851 | unsigned long descriptor_size, | |
852 | u32 descriptor_version, | |
59f2a619 AB |
853 | efi_memory_desc_t *virtual_map, |
854 | unsigned long systab_phys) | |
69829470 | 855 | { |
59f2a619 | 856 | const efi_system_table_t *systab = (efi_system_table_t *)systab_phys; |
69829470 AB |
857 | efi_status_t status; |
858 | unsigned long flags; | |
69829470 | 859 | |
ea5e1919 AB |
860 | if (efi_is_mixed()) |
861 | return efi_thunk_set_virtual_address_map(memory_map_size, | |
862 | descriptor_size, | |
863 | descriptor_version, | |
864 | virtual_map); | |
514b1a84 | 865 | efi_enter_mm(); |
69829470 | 866 | |
b0dc553c | 867 | efi_fpu_begin(); |
e5f930fe | 868 | |
69829470 AB |
869 | /* Disable interrupts around EFI calls: */ |
870 | local_irq_save(flags); | |
0303c972 TG |
871 | status = arch_efi_call_virt(efi.runtime, set_virtual_address_map, |
872 | memory_map_size, descriptor_size, | |
873 | descriptor_version, virtual_map); | |
69829470 AB |
874 | local_irq_restore(flags); |
875 | ||
b0dc553c | 876 | efi_fpu_end(); |
69829470 | 877 | |
59f2a619 AB |
878 | /* grab the virtually remapped EFI runtime services table pointer */ |
879 | efi.runtime = READ_ONCE(systab->runtime); | |
880 | ||
514b1a84 | 881 | efi_leave_mm(); |
69829470 AB |
882 | |
883 | return status; | |
884 | } |