1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * handle transition of Linux booting another kernel
4 * Copyright (C) 2002-2005 Eric Biederman <ebiederm@xmission.com>
5 */
6
7 #define pr_fmt(fmt) "kexec: " fmt
8
9 #include <linux/mm.h>
10 #include <linux/kexec.h>
11 #include <linux/string.h>
12 #include <linux/gfp.h>
13 #include <linux/reboot.h>
14 #include <linux/numa.h>
15 #include <linux/ftrace.h>
16 #include <linux/io.h>
17 #include <linux/suspend.h>
18 #include <linux/vmalloc.h>
19 #include <linux/efi.h>
20 #include <linux/cc_platform.h>
21
22 #include <asm/init.h>
23 #include <asm/tlbflush.h>
24 #include <asm/mmu_context.h>
25 #include <asm/io_apic.h>
26 #include <asm/debugreg.h>
27 #include <asm/kexec-bzimage64.h>
28 #include <asm/setup.h>
29 #include <asm/set_memory.h>
30 #include <asm/cpu.h>
31
32 #ifdef CONFIG_ACPI
33 /*
34 * Used while adding mapping for ACPI tables.
35 * Can be reused when other iomem regions need be mapped
36 */
37 struct init_pgtable_data {
38 struct x86_mapping_info *info;
39 pgd_t *level4p;
40 };
41
mem_region_callback(struct resource * res,void * arg)42 static int mem_region_callback(struct resource *res, void *arg)
43 {
44 struct init_pgtable_data *data = arg;
45 unsigned long mstart, mend;
46
47 mstart = res->start;
48 mend = mstart + resource_size(res) - 1;
49
50 return kernel_ident_mapping_init(data->info, data->level4p, mstart, mend);
51 }
52
53 static int
map_acpi_tables(struct x86_mapping_info * info,pgd_t * level4p)54 map_acpi_tables(struct x86_mapping_info *info, pgd_t *level4p)
55 {
56 struct init_pgtable_data data;
57 unsigned long flags;
58 int ret;
59
60 data.info = info;
61 data.level4p = level4p;
62 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
63
64 ret = walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1,
65 &data, mem_region_callback);
66 if (ret && ret != -EINVAL)
67 return ret;
68
69 /* ACPI tables could be located in ACPI Non-volatile Storage region */
70 ret = walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1,
71 &data, mem_region_callback);
72 if (ret && ret != -EINVAL)
73 return ret;
74
75 return 0;
76 }
77 #else
map_acpi_tables(struct x86_mapping_info * info,pgd_t * level4p)78 static int map_acpi_tables(struct x86_mapping_info *info, pgd_t *level4p) { return 0; }
79 #endif
80
81 #ifdef CONFIG_KEXEC_FILE
82 const struct kexec_file_ops * const kexec_file_loaders[] = {
83 &kexec_bzImage64_ops,
84 NULL
85 };
86 #endif
87
88 static int
map_efi_systab(struct x86_mapping_info * info,pgd_t * level4p)89 map_efi_systab(struct x86_mapping_info *info, pgd_t *level4p)
90 {
91 #ifdef CONFIG_EFI
92 unsigned long mstart, mend;
93
94 if (!efi_enabled(EFI_BOOT))
95 return 0;
96
97 mstart = (boot_params.efi_info.efi_systab |
98 ((u64)boot_params.efi_info.efi_systab_hi<<32));
99
100 if (efi_enabled(EFI_64BIT))
101 mend = mstart + sizeof(efi_system_table_64_t);
102 else
103 mend = mstart + sizeof(efi_system_table_32_t);
104
105 if (!mstart)
106 return 0;
107
108 return kernel_ident_mapping_init(info, level4p, mstart, mend);
109 #endif
110 return 0;
111 }
112
free_transition_pgtable(struct kimage * image)113 static void free_transition_pgtable(struct kimage *image)
114 {
115 free_page((unsigned long)image->arch.p4d);
116 image->arch.p4d = NULL;
117 free_page((unsigned long)image->arch.pud);
118 image->arch.pud = NULL;
119 free_page((unsigned long)image->arch.pmd);
120 image->arch.pmd = NULL;
121 free_page((unsigned long)image->arch.pte);
122 image->arch.pte = NULL;
123 }
124
init_transition_pgtable(struct kimage * image,pgd_t * pgd)125 static int init_transition_pgtable(struct kimage *image, pgd_t *pgd)
126 {
127 pgprot_t prot = PAGE_KERNEL_EXEC_NOENC;
128 unsigned long vaddr, paddr;
129 int result = -ENOMEM;
130 p4d_t *p4d;
131 pud_t *pud;
132 pmd_t *pmd;
133 pte_t *pte;
134
135 vaddr = (unsigned long)relocate_kernel;
136 paddr = __pa(page_address(image->control_code_page)+PAGE_SIZE);
137 pgd += pgd_index(vaddr);
138 if (!pgd_present(*pgd)) {
139 p4d = (p4d_t *)get_zeroed_page(GFP_KERNEL);
140 if (!p4d)
141 goto err;
142 image->arch.p4d = p4d;
143 set_pgd(pgd, __pgd(__pa(p4d) | _KERNPG_TABLE));
144 }
145 p4d = p4d_offset(pgd, vaddr);
146 if (!p4d_present(*p4d)) {
147 pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
148 if (!pud)
149 goto err;
150 image->arch.pud = pud;
151 set_p4d(p4d, __p4d(__pa(pud) | _KERNPG_TABLE));
152 }
153 pud = pud_offset(p4d, vaddr);
154 if (!pud_present(*pud)) {
155 pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
156 if (!pmd)
157 goto err;
158 image->arch.pmd = pmd;
159 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
160 }
161 pmd = pmd_offset(pud, vaddr);
162 if (!pmd_present(*pmd)) {
163 pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
164 if (!pte)
165 goto err;
166 image->arch.pte = pte;
167 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
168 }
169 pte = pte_offset_kernel(pmd, vaddr);
170
171 if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT))
172 prot = PAGE_KERNEL_EXEC;
173
174 set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, prot));
175 return 0;
176 err:
177 return result;
178 }
179
alloc_pgt_page(void * data)180 static void *alloc_pgt_page(void *data)
181 {
182 struct kimage *image = (struct kimage *)data;
183 struct page *page;
184 void *p = NULL;
185
186 page = kimage_alloc_control_pages(image, 0);
187 if (page) {
188 p = page_address(page);
189 clear_page(p);
190 }
191
192 return p;
193 }
194
init_pgtable(struct kimage * image,unsigned long start_pgtable)195 static int init_pgtable(struct kimage *image, unsigned long start_pgtable)
196 {
197 struct x86_mapping_info info = {
198 .alloc_pgt_page = alloc_pgt_page,
199 .context = image,
200 .page_flag = __PAGE_KERNEL_LARGE_EXEC,
201 .kernpg_flag = _KERNPG_TABLE_NOENC,
202 };
203 unsigned long mstart, mend;
204 pgd_t *level4p;
205 int result;
206 int i;
207
208 level4p = (pgd_t *)__va(start_pgtable);
209 clear_page(level4p);
210
211 if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT)) {
212 info.page_flag |= _PAGE_ENC;
213 info.kernpg_flag |= _PAGE_ENC;
214 }
215
216 if (direct_gbpages)
217 info.direct_gbpages = true;
218
219 for (i = 0; i < nr_pfn_mapped; i++) {
220 mstart = pfn_mapped[i].start << PAGE_SHIFT;
221 mend = pfn_mapped[i].end << PAGE_SHIFT;
222
223 result = kernel_ident_mapping_init(&info,
224 level4p, mstart, mend);
225 if (result)
226 return result;
227 }
228
229 /*
230 * segments's mem ranges could be outside 0 ~ max_pfn,
231 * for example when jump back to original kernel from kexeced kernel.
232 * or first kernel is booted with user mem map, and second kernel
233 * could be loaded out of that range.
234 */
235 for (i = 0; i < image->nr_segments; i++) {
236 mstart = image->segment[i].mem;
237 mend = mstart + image->segment[i].memsz;
238
239 result = kernel_ident_mapping_init(&info,
240 level4p, mstart, mend);
241
242 if (result)
243 return result;
244 }
245
246 /*
247 * Prepare EFI systab and ACPI tables for kexec kernel since they are
248 * not covered by pfn_mapped.
249 */
250 result = map_efi_systab(&info, level4p);
251 if (result)
252 return result;
253
254 result = map_acpi_tables(&info, level4p);
255 if (result)
256 return result;
257
258 return init_transition_pgtable(image, level4p);
259 }
260
load_segments(void)261 static void load_segments(void)
262 {
263 __asm__ __volatile__ (
264 "\tmovl %0,%%ds\n"
265 "\tmovl %0,%%es\n"
266 "\tmovl %0,%%ss\n"
267 "\tmovl %0,%%fs\n"
268 "\tmovl %0,%%gs\n"
269 : : "a" (__KERNEL_DS) : "memory"
270 );
271 }
272
machine_kexec_prepare(struct kimage * image)273 int machine_kexec_prepare(struct kimage *image)
274 {
275 unsigned long start_pgtable;
276 int result;
277
278 /* Calculate the offsets */
279 start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
280
281 /* Setup the identity mapped 64bit page table */
282 result = init_pgtable(image, start_pgtable);
283 if (result)
284 return result;
285
286 return 0;
287 }
288
machine_kexec_cleanup(struct kimage * image)289 void machine_kexec_cleanup(struct kimage *image)
290 {
291 free_transition_pgtable(image);
292 }
293
294 /*
295 * Do not allocate memory (or fail in any way) in machine_kexec().
296 * We are past the point of no return, committed to rebooting now.
297 */
machine_kexec(struct kimage * image)298 void machine_kexec(struct kimage *image)
299 {
300 unsigned long page_list[PAGES_NR];
301 unsigned int host_mem_enc_active;
302 int save_ftrace_enabled;
303 void *control_page;
304
305 /*
306 * This must be done before load_segments() since if call depth tracking
307 * is used then GS must be valid to make any function calls.
308 */
309 host_mem_enc_active = cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT);
310
311 #ifdef CONFIG_KEXEC_JUMP
312 if (image->preserve_context)
313 save_processor_state();
314 #endif
315
316 save_ftrace_enabled = __ftrace_enabled_save();
317
318 /* Interrupts aren't acceptable while we reboot */
319 local_irq_disable();
320 hw_breakpoint_disable();
321 cet_disable();
322
323 if (image->preserve_context) {
324 #ifdef CONFIG_X86_IO_APIC
325 /*
326 * We need to put APICs in legacy mode so that we can
327 * get timer interrupts in second kernel. kexec/kdump
328 * paths already have calls to restore_boot_irq_mode()
329 * in one form or other. kexec jump path also need one.
330 */
331 clear_IO_APIC();
332 restore_boot_irq_mode();
333 #endif
334 }
335
336 control_page = page_address(image->control_code_page) + PAGE_SIZE;
337 __memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
338
339 page_list[PA_CONTROL_PAGE] = virt_to_phys(control_page);
340 page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
341 page_list[PA_TABLE_PAGE] =
342 (unsigned long)__pa(page_address(image->control_code_page));
343
344 if (image->type == KEXEC_TYPE_DEFAULT)
345 page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
346 << PAGE_SHIFT);
347
348 /*
349 * The segment registers are funny things, they have both a
350 * visible and an invisible part. Whenever the visible part is
351 * set to a specific selector, the invisible part is loaded
352 * with from a table in memory. At no other time is the
353 * descriptor table in memory accessed.
354 *
355 * I take advantage of this here by force loading the
356 * segments, before I zap the gdt with an invalid value.
357 */
358 load_segments();
359 /*
360 * The gdt & idt are now invalid.
361 * If you want to load them you must set up your own idt & gdt.
362 */
363 native_idt_invalidate();
364 native_gdt_invalidate();
365
366 /* now call it */
367 image->start = relocate_kernel((unsigned long)image->head,
368 (unsigned long)page_list,
369 image->start,
370 image->preserve_context,
371 host_mem_enc_active);
372
373 #ifdef CONFIG_KEXEC_JUMP
374 if (image->preserve_context)
375 restore_processor_state();
376 #endif
377
378 __ftrace_enabled_restore(save_ftrace_enabled);
379 }
380
381 /* arch-dependent functionality related to kexec file-based syscall */
382
383 #ifdef CONFIG_KEXEC_FILE
384 /*
385 * Apply purgatory relocations.
386 *
387 * @pi: Purgatory to be relocated.
388 * @section: Section relocations applying to.
389 * @relsec: Section containing RELAs.
390 * @symtabsec: Corresponding symtab.
391 *
392 * TODO: Some of the code belongs to generic code. Move that in kexec.c.
393 */
arch_kexec_apply_relocations_add(struct purgatory_info * pi,Elf_Shdr * section,const Elf_Shdr * relsec,const Elf_Shdr * symtabsec)394 int arch_kexec_apply_relocations_add(struct purgatory_info *pi,
395 Elf_Shdr *section, const Elf_Shdr *relsec,
396 const Elf_Shdr *symtabsec)
397 {
398 unsigned int i;
399 Elf64_Rela *rel;
400 Elf64_Sym *sym;
401 void *location;
402 unsigned long address, sec_base, value;
403 const char *strtab, *name, *shstrtab;
404 const Elf_Shdr *sechdrs;
405
406 /* String & section header string table */
407 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
408 strtab = (char *)pi->ehdr + sechdrs[symtabsec->sh_link].sh_offset;
409 shstrtab = (char *)pi->ehdr + sechdrs[pi->ehdr->e_shstrndx].sh_offset;
410
411 rel = (void *)pi->ehdr + relsec->sh_offset;
412
413 pr_debug("Applying relocate section %s to %u\n",
414 shstrtab + relsec->sh_name, relsec->sh_info);
415
416 for (i = 0; i < relsec->sh_size / sizeof(*rel); i++) {
417
418 /*
419 * rel[i].r_offset contains byte offset from beginning
420 * of section to the storage unit affected.
421 *
422 * This is location to update. This is temporary buffer
423 * where section is currently loaded. This will finally be
424 * loaded to a different address later, pointed to by
425 * ->sh_addr. kexec takes care of moving it
426 * (kexec_load_segment()).
427 */
428 location = pi->purgatory_buf;
429 location += section->sh_offset;
430 location += rel[i].r_offset;
431
432 /* Final address of the location */
433 address = section->sh_addr + rel[i].r_offset;
434
435 /*
436 * rel[i].r_info contains information about symbol table index
437 * w.r.t which relocation must be made and type of relocation
438 * to apply. ELF64_R_SYM() and ELF64_R_TYPE() macros get
439 * these respectively.
440 */
441 sym = (void *)pi->ehdr + symtabsec->sh_offset;
442 sym += ELF64_R_SYM(rel[i].r_info);
443
444 if (sym->st_name)
445 name = strtab + sym->st_name;
446 else
447 name = shstrtab + sechdrs[sym->st_shndx].sh_name;
448
449 pr_debug("Symbol: %s info: %02x shndx: %02x value=%llx size: %llx\n",
450 name, sym->st_info, sym->st_shndx, sym->st_value,
451 sym->st_size);
452
453 if (sym->st_shndx == SHN_UNDEF) {
454 pr_err("Undefined symbol: %s\n", name);
455 return -ENOEXEC;
456 }
457
458 if (sym->st_shndx == SHN_COMMON) {
459 pr_err("symbol '%s' in common section\n", name);
460 return -ENOEXEC;
461 }
462
463 if (sym->st_shndx == SHN_ABS)
464 sec_base = 0;
465 else if (sym->st_shndx >= pi->ehdr->e_shnum) {
466 pr_err("Invalid section %d for symbol %s\n",
467 sym->st_shndx, name);
468 return -ENOEXEC;
469 } else
470 sec_base = pi->sechdrs[sym->st_shndx].sh_addr;
471
472 value = sym->st_value;
473 value += sec_base;
474 value += rel[i].r_addend;
475
476 switch (ELF64_R_TYPE(rel[i].r_info)) {
477 case R_X86_64_NONE:
478 break;
479 case R_X86_64_64:
480 *(u64 *)location = value;
481 break;
482 case R_X86_64_32:
483 *(u32 *)location = value;
484 if (value != *(u32 *)location)
485 goto overflow;
486 break;
487 case R_X86_64_32S:
488 *(s32 *)location = value;
489 if ((s64)value != *(s32 *)location)
490 goto overflow;
491 break;
492 case R_X86_64_PC32:
493 case R_X86_64_PLT32:
494 value -= (u64)address;
495 *(u32 *)location = value;
496 break;
497 default:
498 pr_err("Unknown rela relocation: %llu\n",
499 ELF64_R_TYPE(rel[i].r_info));
500 return -ENOEXEC;
501 }
502 }
503 return 0;
504
505 overflow:
506 pr_err("Overflow in relocation type %d value 0x%lx\n",
507 (int)ELF64_R_TYPE(rel[i].r_info), value);
508 return -ENOEXEC;
509 }
510
arch_kimage_file_post_load_cleanup(struct kimage * image)511 int arch_kimage_file_post_load_cleanup(struct kimage *image)
512 {
513 vfree(image->elf_headers);
514 image->elf_headers = NULL;
515 image->elf_headers_sz = 0;
516
517 return kexec_image_post_load_cleanup_default(image);
518 }
519 #endif /* CONFIG_KEXEC_FILE */
520
521 static int
kexec_mark_range(unsigned long start,unsigned long end,bool protect)522 kexec_mark_range(unsigned long start, unsigned long end, bool protect)
523 {
524 struct page *page;
525 unsigned int nr_pages;
526
527 /*
528 * For physical range: [start, end]. We must skip the unassigned
529 * crashk resource with zero-valued "end" member.
530 */
531 if (!end || start > end)
532 return 0;
533
534 page = pfn_to_page(start >> PAGE_SHIFT);
535 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
536 if (protect)
537 return set_pages_ro(page, nr_pages);
538 else
539 return set_pages_rw(page, nr_pages);
540 }
541
kexec_mark_crashkres(bool protect)542 static void kexec_mark_crashkres(bool protect)
543 {
544 unsigned long control;
545
546 kexec_mark_range(crashk_low_res.start, crashk_low_res.end, protect);
547
548 /* Don't touch the control code page used in crash_kexec().*/
549 control = PFN_PHYS(page_to_pfn(kexec_crash_image->control_code_page));
550 /* Control code page is located in the 2nd page. */
551 kexec_mark_range(crashk_res.start, control + PAGE_SIZE - 1, protect);
552 control += KEXEC_CONTROL_PAGE_SIZE;
553 kexec_mark_range(control, crashk_res.end, protect);
554 }
555
arch_kexec_protect_crashkres(void)556 void arch_kexec_protect_crashkres(void)
557 {
558 kexec_mark_crashkres(true);
559 }
560
arch_kexec_unprotect_crashkres(void)561 void arch_kexec_unprotect_crashkres(void)
562 {
563 kexec_mark_crashkres(false);
564 }
565
566 /*
567 * During a traditional boot under SME, SME will encrypt the kernel,
568 * so the SME kexec kernel also needs to be un-encrypted in order to
569 * replicate a normal SME boot.
570 *
571 * During a traditional boot under SEV, the kernel has already been
572 * loaded encrypted, so the SEV kexec kernel needs to be encrypted in
573 * order to replicate a normal SEV boot.
574 */
arch_kexec_post_alloc_pages(void * vaddr,unsigned int pages,gfp_t gfp)575 int arch_kexec_post_alloc_pages(void *vaddr, unsigned int pages, gfp_t gfp)
576 {
577 if (!cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT))
578 return 0;
579
580 /*
581 * If host memory encryption is active we need to be sure that kexec
582 * pages are not encrypted because when we boot to the new kernel the
583 * pages won't be accessed encrypted (initially).
584 */
585 return set_memory_decrypted((unsigned long)vaddr, pages);
586 }
587
arch_kexec_pre_free_pages(void * vaddr,unsigned int pages)588 void arch_kexec_pre_free_pages(void *vaddr, unsigned int pages)
589 {
590 if (!cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT))
591 return;
592
593 /*
594 * If host memory encryption is active we need to reset the pages back
595 * to being an encrypted mapping before freeing them.
596 */
597 set_memory_encrypted((unsigned long)vaddr, pages);
598 }
599