1 // SPDX-License-Identifier: GPL-2.0-only
3 * handle transition of Linux booting another kernel
4 * Copyright (C) 2002-2005 Eric Biederman <ebiederm@xmission.com>
7 #define pr_fmt(fmt) "kexec: " fmt
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>
17 #include <linux/suspend.h>
18 #include <linux/vmalloc.h>
19 #include <linux/efi.h>
20 #include <linux/cc_platform.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>
34 * Used while adding mapping for ACPI tables.
35 * Can be reused when other iomem regions need be mapped
37 struct init_pgtable_data {
38 struct x86_mapping_info *info;
42 static int mem_region_callback(struct resource *res, void *arg)
44 struct init_pgtable_data *data = arg;
46 return kernel_ident_mapping_init(data->info, data->level4p,
47 res->start, res->end + 1);
51 map_acpi_tables(struct x86_mapping_info *info, pgd_t *level4p)
53 struct init_pgtable_data data;
58 data.level4p = level4p;
59 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
61 ret = walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1,
62 &data, mem_region_callback);
63 if (ret && ret != -EINVAL)
66 /* ACPI tables could be located in ACPI Non-volatile Storage region */
67 ret = walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1,
68 &data, mem_region_callback);
69 if (ret && ret != -EINVAL)
75 static int map_acpi_tables(struct x86_mapping_info *info, pgd_t *level4p) { return 0; }
78 #ifdef CONFIG_KEXEC_FILE
79 const struct kexec_file_ops * const kexec_file_loaders[] = {
86 map_efi_systab(struct x86_mapping_info *info, pgd_t *level4p)
89 unsigned long mstart, mend;
91 if (!efi_enabled(EFI_BOOT))
94 mstart = (boot_params.efi_info.efi_systab |
95 ((u64)boot_params.efi_info.efi_systab_hi<<32));
97 if (efi_enabled(EFI_64BIT))
98 mend = mstart + sizeof(efi_system_table_64_t);
100 mend = mstart + sizeof(efi_system_table_32_t);
105 return kernel_ident_mapping_init(info, level4p, mstart, mend);
110 static void free_transition_pgtable(struct kimage *image)
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;
122 static int init_transition_pgtable(struct kimage *image, pgd_t *pgd)
124 pgprot_t prot = PAGE_KERNEL_EXEC_NOENC;
125 unsigned long vaddr, paddr;
126 int result = -ENOMEM;
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)) {
136 p4d = (p4d_t *)get_zeroed_page(GFP_KERNEL);
139 image->arch.p4d = p4d;
140 set_pgd(pgd, __pgd(__pa(p4d) | _KERNPG_TABLE));
142 p4d = p4d_offset(pgd, vaddr);
143 if (!p4d_present(*p4d)) {
144 pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
147 image->arch.pud = pud;
148 set_p4d(p4d, __p4d(__pa(pud) | _KERNPG_TABLE));
150 pud = pud_offset(p4d, vaddr);
151 if (!pud_present(*pud)) {
152 pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
155 image->arch.pmd = pmd;
156 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
158 pmd = pmd_offset(pud, vaddr);
159 if (!pmd_present(*pmd)) {
160 pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
163 image->arch.pte = pte;
164 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
166 pte = pte_offset_kernel(pmd, vaddr);
168 if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT))
169 prot = PAGE_KERNEL_EXEC;
171 set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, prot));
177 static void *alloc_pgt_page(void *data)
179 struct kimage *image = (struct kimage *)data;
183 page = kimage_alloc_control_pages(image, 0);
185 p = page_address(page);
192 static int init_pgtable(struct kimage *image, unsigned long start_pgtable)
194 struct x86_mapping_info info = {
195 .alloc_pgt_page = alloc_pgt_page,
197 .page_flag = __PAGE_KERNEL_LARGE_EXEC,
198 .kernpg_flag = _KERNPG_TABLE_NOENC,
200 unsigned long mstart, mend;
205 level4p = (pgd_t *)__va(start_pgtable);
208 if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT)) {
209 info.page_flag |= _PAGE_ENC;
210 info.kernpg_flag |= _PAGE_ENC;
214 info.direct_gbpages = true;
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;
220 result = kernel_ident_mapping_init(&info,
221 level4p, mstart, mend);
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.
232 for (i = 0; i < image->nr_segments; i++) {
233 mstart = image->segment[i].mem;
234 mend = mstart + image->segment[i].memsz;
236 result = kernel_ident_mapping_init(&info,
237 level4p, mstart, mend);
244 * Prepare EFI systab and ACPI tables for kexec kernel since they are
245 * not covered by pfn_mapped.
247 result = map_efi_systab(&info, level4p);
251 result = map_acpi_tables(&info, level4p);
255 return init_transition_pgtable(image, level4p);
258 static void load_segments(void)
260 __asm__ __volatile__ (
266 : : "a" (__KERNEL_DS) : "memory"
270 int machine_kexec_prepare(struct kimage *image)
272 unsigned long start_pgtable;
275 /* Calculate the offsets */
276 start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
278 /* Setup the identity mapped 64bit page table */
279 result = init_pgtable(image, start_pgtable);
286 void machine_kexec_cleanup(struct kimage *image)
288 free_transition_pgtable(image);
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.
295 void machine_kexec(struct kimage *image)
297 unsigned long page_list[PAGES_NR];
299 int save_ftrace_enabled;
301 #ifdef CONFIG_KEXEC_JUMP
302 if (image->preserve_context)
303 save_processor_state();
306 save_ftrace_enabled = __ftrace_enabled_save();
308 /* Interrupts aren't acceptable while we reboot */
310 hw_breakpoint_disable();
313 if (image->preserve_context) {
314 #ifdef CONFIG_X86_IO_APIC
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.
322 restore_boot_irq_mode();
326 control_page = page_address(image->control_code_page) + PAGE_SIZE;
327 __memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
329 page_list[PA_CONTROL_PAGE] = virt_to_phys(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));
334 if (image->type == KEXEC_TYPE_DEFAULT)
335 page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
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.
345 * I take advantage of this here by force loading the
346 * segments, before I zap the gdt with an invalid value.
350 * The gdt & idt are now invalid.
351 * If you want to load them you must set up your own idt & gdt.
353 native_idt_invalidate();
354 native_gdt_invalidate();
357 image->start = relocate_kernel((unsigned long)image->head,
358 (unsigned long)page_list,
360 image->preserve_context,
361 cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT));
363 #ifdef CONFIG_KEXEC_JUMP
364 if (image->preserve_context)
365 restore_processor_state();
368 __ftrace_enabled_restore(save_ftrace_enabled);
371 /* arch-dependent functionality related to kexec file-based syscall */
373 #ifdef CONFIG_KEXEC_FILE
375 * Apply purgatory relocations.
377 * @pi: Purgatory to be relocated.
378 * @section: Section relocations applying to.
379 * @relsec: Section containing RELAs.
380 * @symtabsec: Corresponding symtab.
382 * TODO: Some of the code belongs to generic code. Move that in kexec.c.
384 int arch_kexec_apply_relocations_add(struct purgatory_info *pi,
385 Elf_Shdr *section, const Elf_Shdr *relsec,
386 const Elf_Shdr *symtabsec)
392 unsigned long address, sec_base, value;
393 const char *strtab, *name, *shstrtab;
394 const Elf_Shdr *sechdrs;
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;
401 rel = (void *)pi->ehdr + relsec->sh_offset;
403 pr_debug("Applying relocate section %s to %u\n",
404 shstrtab + relsec->sh_name, relsec->sh_info);
406 for (i = 0; i < relsec->sh_size / sizeof(*rel); i++) {
409 * rel[i].r_offset contains byte offset from beginning
410 * of section to the storage unit affected.
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()).
418 location = pi->purgatory_buf;
419 location += section->sh_offset;
420 location += rel[i].r_offset;
422 /* Final address of the location */
423 address = section->sh_addr + rel[i].r_offset;
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.
431 sym = (void *)pi->ehdr + symtabsec->sh_offset;
432 sym += ELF64_R_SYM(rel[i].r_info);
435 name = strtab + sym->st_name;
437 name = shstrtab + sechdrs[sym->st_shndx].sh_name;
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,
443 if (sym->st_shndx == SHN_UNDEF) {
444 pr_err("Undefined symbol: %s\n", name);
448 if (sym->st_shndx == SHN_COMMON) {
449 pr_err("symbol '%s' in common section\n", name);
453 if (sym->st_shndx == SHN_ABS)
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);
460 sec_base = pi->sechdrs[sym->st_shndx].sh_addr;
462 value = sym->st_value;
464 value += rel[i].r_addend;
466 switch (ELF64_R_TYPE(rel[i].r_info)) {
470 *(u64 *)location = value;
473 *(u32 *)location = value;
474 if (value != *(u32 *)location)
478 *(s32 *)location = value;
479 if ((s64)value != *(s32 *)location)
484 value -= (u64)address;
485 *(u32 *)location = value;
488 pr_err("Unknown rela relocation: %llu\n",
489 ELF64_R_TYPE(rel[i].r_info));
496 pr_err("Overflow in relocation type %d value 0x%lx\n",
497 (int)ELF64_R_TYPE(rel[i].r_info), value);
501 int arch_kimage_file_post_load_cleanup(struct kimage *image)
503 vfree(image->elf_headers);
504 image->elf_headers = NULL;
505 image->elf_headers_sz = 0;
507 return kexec_image_post_load_cleanup_default(image);
509 #endif /* CONFIG_KEXEC_FILE */
511 #ifdef CONFIG_CRASH_DUMP
514 kexec_mark_range(unsigned long start, unsigned long end, bool protect)
517 unsigned int nr_pages;
520 * For physical range: [start, end]. We must skip the unassigned
521 * crashk resource with zero-valued "end" member.
523 if (!end || start > end)
526 page = pfn_to_page(start >> PAGE_SHIFT);
527 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
529 return set_pages_ro(page, nr_pages);
531 return set_pages_rw(page, nr_pages);
534 static void kexec_mark_crashkres(bool protect)
536 unsigned long control;
538 kexec_mark_range(crashk_low_res.start, crashk_low_res.end, protect);
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(crashk_res.start, control + PAGE_SIZE - 1, protect);
544 control += KEXEC_CONTROL_PAGE_SIZE;
545 kexec_mark_range(control, crashk_res.end, protect);
548 void arch_kexec_protect_crashkres(void)
550 kexec_mark_crashkres(true);
553 void arch_kexec_unprotect_crashkres(void)
555 kexec_mark_crashkres(false);
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.
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.
568 int arch_kexec_post_alloc_pages(void *vaddr, unsigned int pages, gfp_t gfp)
570 if (!cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT))
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).
578 return set_memory_decrypted((unsigned long)vaddr, pages);
581 void arch_kexec_pre_free_pages(void *vaddr, unsigned int pages)
583 if (!cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT))
587 * If host memory encryption is active we need to reset the pages back
588 * to being an encrypted mapping before freeing them.
590 set_memory_encrypted((unsigned long)vaddr, pages);