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 */
37struct init_pgtable_data {
38 struct x86_mapping_info *info;
39 pgd_t *level4p;
40};
41
42static int mem_region_callback(struct resource *res, void *arg)
43{
44 struct init_pgtable_data *data = arg;
45
46 return kernel_ident_mapping_init(info: data->info, pgd_page: data->level4p,
47 pstart: res->start, pend: res->end + 1);
48}
49
50static int
51map_acpi_tables(struct x86_mapping_info *info, pgd_t *level4p)
52{
53 struct init_pgtable_data data;
54 unsigned long flags;
55 int ret;
56
57 data.info = info;
58 data.level4p = level4p;
59 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
60
61 ret = walk_iomem_res_desc(desc: IORES_DESC_ACPI_TABLES, flags, start: 0, end: -1,
62 arg: &data, func: mem_region_callback);
63 if (ret && ret != -EINVAL)
64 return ret;
65
66 /* ACPI tables could be located in ACPI Non-volatile Storage region */
67 ret = walk_iomem_res_desc(desc: IORES_DESC_ACPI_NV_STORAGE, flags, start: 0, end: -1,
68 arg: &data, func: mem_region_callback);
69 if (ret && ret != -EINVAL)
70 return ret;
71
72 return 0;
73}
74#else
75static int map_acpi_tables(struct x86_mapping_info *info, pgd_t *level4p) { return 0; }
76#endif
77
78#ifdef CONFIG_KEXEC_FILE
79const struct kexec_file_ops * const kexec_file_loaders[] = {
80 &kexec_bzImage64_ops,
81 NULL
82};
83#endif
84
85static int
86map_efi_systab(struct x86_mapping_info *info, pgd_t *level4p)
87{
88#ifdef CONFIG_EFI
89 unsigned long mstart, mend;
90
91 if (!efi_enabled(EFI_BOOT))
92 return 0;
93
94 mstart = (boot_params.efi_info.efi_systab |
95 ((u64)boot_params.efi_info.efi_systab_hi<<32));
96
97 if (efi_enabled(EFI_64BIT))
98 mend = mstart + sizeof(efi_system_table_64_t);
99 else
100 mend = mstart + sizeof(efi_system_table_32_t);
101
102 if (!mstart)
103 return 0;
104
105 return kernel_ident_mapping_init(info, pgd_page: level4p, pstart: mstart, pend: mend);
106#endif
107 return 0;
108}
109
110static void free_transition_pgtable(struct kimage *image)
111{
112 free_page((unsigned long)image->arch.p4d);
113 image->arch.p4d = NULL;
114 free_page((unsigned long)image->arch.pud);
115 image->arch.pud = NULL;
116 free_page((unsigned long)image->arch.pmd);
117 image->arch.pmd = NULL;
118 free_page((unsigned long)image->arch.pte);
119 image->arch.pte = NULL;
120}
121
122static int init_transition_pgtable(struct kimage *image, pgd_t *pgd)
123{
124 pgprot_t prot = PAGE_KERNEL_EXEC_NOENC;
125 unsigned long vaddr, paddr;
126 int result = -ENOMEM;
127 p4d_t *p4d;
128 pud_t *pud;
129 pmd_t *pmd;
130 pte_t *pte;
131
132 vaddr = (unsigned long)relocate_kernel;
133 paddr = __pa(page_address(image->control_code_page)+PAGE_SIZE);
134 pgd += pgd_index(vaddr);
135 if (!pgd_present(pgd: *pgd)) {
136 p4d = (p4d_t *)get_zeroed_page(GFP_KERNEL);
137 if (!p4d)
138 goto err;
139 image->arch.p4d = p4d;
140 set_pgd(pgd, __pgd(__pa(p4d) | _KERNPG_TABLE));
141 }
142 p4d = p4d_offset(pgd, address: vaddr);
143 if (!p4d_present(p4d: *p4d)) {
144 pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
145 if (!pud)
146 goto err;
147 image->arch.pud = pud;
148 set_p4d(p4dp: p4d, p4d: __p4d(__pa(pud) | _KERNPG_TABLE));
149 }
150 pud = pud_offset(p4d, address: vaddr);
151 if (!pud_present(pud: *pud)) {
152 pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
153 if (!pmd)
154 goto err;
155 image->arch.pmd = pmd;
156 set_pud(pudp: pud, pud: __pud(__pa(pmd) | _KERNPG_TABLE));
157 }
158 pmd = pmd_offset(pud, address: vaddr);
159 if (!pmd_present(pmd: *pmd)) {
160 pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
161 if (!pte)
162 goto err;
163 image->arch.pte = pte;
164 set_pmd(pmdp: pmd, pmd: __pmd(__pa(pte) | _KERNPG_TABLE));
165 }
166 pte = pte_offset_kernel(pmd, address: vaddr);
167
168 if (cc_platform_has(attr: CC_ATTR_GUEST_MEM_ENCRYPT))
169 prot = PAGE_KERNEL_EXEC;
170
171 set_pte(ptep: pte, pte: pfn_pte(page_nr: paddr >> PAGE_SHIFT, pgprot: prot));
172 return 0;
173err:
174 return result;
175}
176
177static void *alloc_pgt_page(void *data)
178{
179 struct kimage *image = (struct kimage *)data;
180 struct page *page;
181 void *p = NULL;
182
183 page = kimage_alloc_control_pages(image, order: 0);
184 if (page) {
185 p = page_address(page);
186 clear_page(page: p);
187 }
188
189 return p;
190}
191
192static int init_pgtable(struct kimage *image, unsigned long start_pgtable)
193{
194 struct x86_mapping_info info = {
195 .alloc_pgt_page = alloc_pgt_page,
196 .context = image,
197 .page_flag = __PAGE_KERNEL_LARGE_EXEC,
198 .kernpg_flag = _KERNPG_TABLE_NOENC,
199 };
200 unsigned long mstart, mend;
201 pgd_t *level4p;
202 int result;
203 int i;
204
205 level4p = (pgd_t *)__va(start_pgtable);
206 clear_page(page: level4p);
207
208 if (cc_platform_has(attr: CC_ATTR_GUEST_MEM_ENCRYPT)) {
209 info.page_flag |= _PAGE_ENC;
210 info.kernpg_flag |= _PAGE_ENC;
211 }
212
213 if (direct_gbpages)
214 info.direct_gbpages = true;
215
216 for (i = 0; i < nr_pfn_mapped; i++) {
217 mstart = pfn_mapped[i].start << PAGE_SHIFT;
218 mend = pfn_mapped[i].end << PAGE_SHIFT;
219
220 result = kernel_ident_mapping_init(info: &info,
221 pgd_page: level4p, pstart: mstart, pend: mend);
222 if (result)
223 return result;
224 }
225
226 /*
227 * segments's mem ranges could be outside 0 ~ max_pfn,
228 * for example when jump back to original kernel from kexeced kernel.
229 * or first kernel is booted with user mem map, and second kernel
230 * could be loaded out of that range.
231 */
232 for (i = 0; i < image->nr_segments; i++) {
233 mstart = image->segment[i].mem;
234 mend = mstart + image->segment[i].memsz;
235
236 result = kernel_ident_mapping_init(info: &info,
237 pgd_page: level4p, pstart: mstart, pend: mend);
238
239 if (result)
240 return result;
241 }
242
243 /*
244 * Prepare EFI systab and ACPI tables for kexec kernel since they are
245 * not covered by pfn_mapped.
246 */
247 result = map_efi_systab(info: &info, level4p);
248 if (result)
249 return result;
250
251 result = map_acpi_tables(info: &info, level4p);
252 if (result)
253 return result;
254
255 return init_transition_pgtable(image, pgd: level4p);
256}
257
258static void load_segments(void)
259{
260 __asm__ __volatile__ (
261 "\tmovl %0,%%ds\n"
262 "\tmovl %0,%%es\n"
263 "\tmovl %0,%%ss\n"
264 "\tmovl %0,%%fs\n"
265 "\tmovl %0,%%gs\n"
266 : : "a" (__KERNEL_DS) : "memory"
267 );
268}
269
270int machine_kexec_prepare(struct kimage *image)
271{
272 unsigned long start_pgtable;
273 int result;
274
275 /* Calculate the offsets */
276 start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
277
278 /* Setup the identity mapped 64bit page table */
279 result = init_pgtable(image, start_pgtable);
280 if (result)
281 return result;
282
283 return 0;
284}
285
286void machine_kexec_cleanup(struct kimage *image)
287{
288 free_transition_pgtable(image);
289}
290
291/*
292 * Do not allocate memory (or fail in any way) in machine_kexec().
293 * We are past the point of no return, committed to rebooting now.
294 */
295void machine_kexec(struct kimage *image)
296{
297 unsigned long page_list[PAGES_NR];
298 void *control_page;
299 int save_ftrace_enabled;
300
301#ifdef CONFIG_KEXEC_JUMP
302 if (image->preserve_context)
303 save_processor_state();
304#endif
305
306 save_ftrace_enabled = __ftrace_enabled_save();
307
308 /* Interrupts aren't acceptable while we reboot */
309 local_irq_disable();
310 hw_breakpoint_disable();
311 cet_disable();
312
313 if (image->preserve_context) {
314#ifdef CONFIG_X86_IO_APIC
315 /*
316 * We need to put APICs in legacy mode so that we can
317 * get timer interrupts in second kernel. kexec/kdump
318 * paths already have calls to restore_boot_irq_mode()
319 * in one form or other. kexec jump path also need one.
320 */
321 clear_IO_APIC();
322 restore_boot_irq_mode();
323#endif
324 }
325
326 control_page = page_address(image->control_code_page) + PAGE_SIZE;
327 __memcpy(to: control_page, from: relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
328
329 page_list[PA_CONTROL_PAGE] = virt_to_phys(address: control_page);
330 page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
331 page_list[PA_TABLE_PAGE] =
332 (unsigned long)__pa(page_address(image->control_code_page));
333
334 if (image->type == KEXEC_TYPE_DEFAULT)
335 page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
336 << PAGE_SHIFT);
337
338 /*
339 * The segment registers are funny things, they have both a
340 * visible and an invisible part. Whenever the visible part is
341 * set to a specific selector, the invisible part is loaded
342 * with from a table in memory. At no other time is the
343 * descriptor table in memory accessed.
344 *
345 * I take advantage of this here by force loading the
346 * segments, before I zap the gdt with an invalid value.
347 */
348 load_segments();
349 /*
350 * The gdt & idt are now invalid.
351 * If you want to load them you must set up your own idt & gdt.
352 */
353 native_idt_invalidate();
354 native_gdt_invalidate();
355
356 /* now call it */
357 image->start = relocate_kernel(indirection_page: (unsigned long)image->head,
358 page_list: (unsigned long)page_list,
359 start_address: image->start,
360 preserve_context: image->preserve_context,
361 host_mem_enc_active: cc_platform_has(attr: CC_ATTR_HOST_MEM_ENCRYPT));
362
363#ifdef CONFIG_KEXEC_JUMP
364 if (image->preserve_context)
365 restore_processor_state();
366#endif
367
368 __ftrace_enabled_restore(enabled: save_ftrace_enabled);
369}
370
371/* arch-dependent functionality related to kexec file-based syscall */
372
373#ifdef CONFIG_KEXEC_FILE
374/*
375 * Apply purgatory relocations.
376 *
377 * @pi: Purgatory to be relocated.
378 * @section: Section relocations applying to.
379 * @relsec: Section containing RELAs.
380 * @symtabsec: Corresponding symtab.
381 *
382 * TODO: Some of the code belongs to generic code. Move that in kexec.c.
383 */
384int arch_kexec_apply_relocations_add(struct purgatory_info *pi,
385 Elf_Shdr *section, const Elf_Shdr *relsec,
386 const Elf_Shdr *symtabsec)
387{
388 unsigned int i;
389 Elf64_Rela *rel;
390 Elf64_Sym *sym;
391 void *location;
392 unsigned long address, sec_base, value;
393 const char *strtab, *name, *shstrtab;
394 const Elf_Shdr *sechdrs;
395
396 /* String & section header string table */
397 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
398 strtab = (char *)pi->ehdr + sechdrs[symtabsec->sh_link].sh_offset;
399 shstrtab = (char *)pi->ehdr + sechdrs[pi->ehdr->e_shstrndx].sh_offset;
400
401 rel = (void *)pi->ehdr + relsec->sh_offset;
402
403 pr_debug("Applying relocate section %s to %u\n",
404 shstrtab + relsec->sh_name, relsec->sh_info);
405
406 for (i = 0; i < relsec->sh_size / sizeof(*rel); i++) {
407
408 /*
409 * rel[i].r_offset contains byte offset from beginning
410 * of section to the storage unit affected.
411 *
412 * This is location to update. This is temporary buffer
413 * where section is currently loaded. This will finally be
414 * loaded to a different address later, pointed to by
415 * ->sh_addr. kexec takes care of moving it
416 * (kexec_load_segment()).
417 */
418 location = pi->purgatory_buf;
419 location += section->sh_offset;
420 location += rel[i].r_offset;
421
422 /* Final address of the location */
423 address = section->sh_addr + rel[i].r_offset;
424
425 /*
426 * rel[i].r_info contains information about symbol table index
427 * w.r.t which relocation must be made and type of relocation
428 * to apply. ELF64_R_SYM() and ELF64_R_TYPE() macros get
429 * these respectively.
430 */
431 sym = (void *)pi->ehdr + symtabsec->sh_offset;
432 sym += ELF64_R_SYM(rel[i].r_info);
433
434 if (sym->st_name)
435 name = strtab + sym->st_name;
436 else
437 name = shstrtab + sechdrs[sym->st_shndx].sh_name;
438
439 pr_debug("Symbol: %s info: %02x shndx: %02x value=%llx size: %llx\n",
440 name, sym->st_info, sym->st_shndx, sym->st_value,
441 sym->st_size);
442
443 if (sym->st_shndx == SHN_UNDEF) {
444 pr_err("Undefined symbol: %s\n", name);
445 return -ENOEXEC;
446 }
447
448 if (sym->st_shndx == SHN_COMMON) {
449 pr_err("symbol '%s' in common section\n", name);
450 return -ENOEXEC;
451 }
452
453 if (sym->st_shndx == SHN_ABS)
454 sec_base = 0;
455 else if (sym->st_shndx >= pi->ehdr->e_shnum) {
456 pr_err("Invalid section %d for symbol %s\n",
457 sym->st_shndx, name);
458 return -ENOEXEC;
459 } else
460 sec_base = pi->sechdrs[sym->st_shndx].sh_addr;
461
462 value = sym->st_value;
463 value += sec_base;
464 value += rel[i].r_addend;
465
466 switch (ELF64_R_TYPE(rel[i].r_info)) {
467 case R_X86_64_NONE:
468 break;
469 case R_X86_64_64:
470 *(u64 *)location = value;
471 break;
472 case R_X86_64_32:
473 *(u32 *)location = value;
474 if (value != *(u32 *)location)
475 goto overflow;
476 break;
477 case R_X86_64_32S:
478 *(s32 *)location = value;
479 if ((s64)value != *(s32 *)location)
480 goto overflow;
481 break;
482 case R_X86_64_PC32:
483 case R_X86_64_PLT32:
484 value -= (u64)address;
485 *(u32 *)location = value;
486 break;
487 default:
488 pr_err("Unknown rela relocation: %llu\n",
489 ELF64_R_TYPE(rel[i].r_info));
490 return -ENOEXEC;
491 }
492 }
493 return 0;
494
495overflow:
496 pr_err("Overflow in relocation type %d value 0x%lx\n",
497 (int)ELF64_R_TYPE(rel[i].r_info), value);
498 return -ENOEXEC;
499}
500
501int arch_kimage_file_post_load_cleanup(struct kimage *image)
502{
503 vfree(addr: image->elf_headers);
504 image->elf_headers = NULL;
505 image->elf_headers_sz = 0;
506
507 return kexec_image_post_load_cleanup_default(image);
508}
509#endif /* CONFIG_KEXEC_FILE */
510
511#ifdef CONFIG_CRASH_DUMP
512
513static int
514kexec_mark_range(unsigned long start, unsigned long end, bool protect)
515{
516 struct page *page;
517 unsigned int nr_pages;
518
519 /*
520 * For physical range: [start, end]. We must skip the unassigned
521 * crashk resource with zero-valued "end" member.
522 */
523 if (!end || start > end)
524 return 0;
525
526 page = pfn_to_page(start >> PAGE_SHIFT);
527 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
528 if (protect)
529 return set_pages_ro(page, numpages: nr_pages);
530 else
531 return set_pages_rw(page, numpages: nr_pages);
532}
533
534static void kexec_mark_crashkres(bool protect)
535{
536 unsigned long control;
537
538 kexec_mark_range(start: crashk_low_res.start, end: crashk_low_res.end, protect);
539
540 /* Don't touch the control code page used in crash_kexec().*/
541 control = PFN_PHYS(page_to_pfn(kexec_crash_image->control_code_page));
542 /* Control code page is located in the 2nd page. */
543 kexec_mark_range(start: crashk_res.start, end: control + PAGE_SIZE - 1, protect);
544 control += KEXEC_CONTROL_PAGE_SIZE;
545 kexec_mark_range(start: control, end: crashk_res.end, protect);
546}
547
548void arch_kexec_protect_crashkres(void)
549{
550 kexec_mark_crashkres(protect: true);
551}
552
553void arch_kexec_unprotect_crashkres(void)
554{
555 kexec_mark_crashkres(protect: false);
556}
557#endif
558
559/*
560 * During a traditional boot under SME, SME will encrypt the kernel,
561 * so the SME kexec kernel also needs to be un-encrypted in order to
562 * replicate a normal SME boot.
563 *
564 * During a traditional boot under SEV, the kernel has already been
565 * loaded encrypted, so the SEV kexec kernel needs to be encrypted in
566 * order to replicate a normal SEV boot.
567 */
568int arch_kexec_post_alloc_pages(void *vaddr, unsigned int pages, gfp_t gfp)
569{
570 if (!cc_platform_has(attr: CC_ATTR_HOST_MEM_ENCRYPT))
571 return 0;
572
573 /*
574 * If host memory encryption is active we need to be sure that kexec
575 * pages are not encrypted because when we boot to the new kernel the
576 * pages won't be accessed encrypted (initially).
577 */
578 return set_memory_decrypted(addr: (unsigned long)vaddr, numpages: pages);
579}
580
581void arch_kexec_pre_free_pages(void *vaddr, unsigned int pages)
582{
583 if (!cc_platform_has(attr: CC_ATTR_HOST_MEM_ENCRYPT))
584 return;
585
586 /*
587 * If host memory encryption is active we need to reset the pages back
588 * to being an encrypted mapping before freeing them.
589 */
590 set_memory_encrypted(addr: (unsigned long)vaddr, numpages: pages);
591}
592

source code of linux/arch/x86/kernel/machine_kexec_64.c