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>
22 #include <asm/pgtable.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>
33 * Used while adding mapping for ACPI tables.
34 * Can be reused when other iomem regions need be mapped
36 struct init_pgtable_data {
37 struct x86_mapping_info *info;
41 static int mem_region_callback(struct resource *res, void *arg)
43 struct init_pgtable_data *data = arg;
44 unsigned long mstart, mend;
47 mend = mstart + resource_size(res) - 1;
49 return kernel_ident_mapping_init(data->info, data->level4p, mstart, mend);
53 map_acpi_tables(struct x86_mapping_info *info, pgd_t *level4p)
55 struct init_pgtable_data data;
60 data.level4p = level4p;
61 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
63 ret = walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1,
64 &data, mem_region_callback);
65 if (ret && ret != -EINVAL)
68 /* ACPI tables could be located in ACPI Non-volatile Storage region */
69 ret = walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1,
70 &data, mem_region_callback);
71 if (ret && ret != -EINVAL)
77 static int map_acpi_tables(struct x86_mapping_info *info, pgd_t *level4p) { return 0; }
80 #ifdef CONFIG_KEXEC_FILE
81 const struct kexec_file_ops * const kexec_file_loaders[] = {
88 map_efi_systab(struct x86_mapping_info *info, pgd_t *level4p)
91 unsigned long mstart, mend;
93 if (!efi_enabled(EFI_BOOT))
96 mstart = (boot_params.efi_info.efi_systab |
97 ((u64)boot_params.efi_info.efi_systab_hi<<32));
99 if (efi_enabled(EFI_64BIT))
100 mend = mstart + sizeof(efi_system_table_64_t);
102 mend = mstart + sizeof(efi_system_table_32_t);
107 return kernel_ident_mapping_init(info, level4p, mstart, mend);
112 static void free_transition_pgtable(struct kimage *image)
114 free_page((unsigned long)image->arch.p4d);
115 image->arch.p4d = NULL;
116 free_page((unsigned long)image->arch.pud);
117 image->arch.pud = NULL;
118 free_page((unsigned long)image->arch.pmd);
119 image->arch.pmd = NULL;
120 free_page((unsigned long)image->arch.pte);
121 image->arch.pte = NULL;
124 static int init_transition_pgtable(struct kimage *image, pgd_t *pgd)
126 pgprot_t prot = PAGE_KERNEL_EXEC_NOENC;
127 unsigned long vaddr, paddr;
128 int result = -ENOMEM;
134 vaddr = (unsigned long)relocate_kernel;
135 paddr = __pa(page_address(image->control_code_page)+PAGE_SIZE);
136 pgd += pgd_index(vaddr);
137 if (!pgd_present(*pgd)) {
138 p4d = (p4d_t *)get_zeroed_page(GFP_KERNEL);
141 image->arch.p4d = p4d;
142 set_pgd(pgd, __pgd(__pa(p4d) | _KERNPG_TABLE));
144 p4d = p4d_offset(pgd, vaddr);
145 if (!p4d_present(*p4d)) {
146 pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
149 image->arch.pud = pud;
150 set_p4d(p4d, __p4d(__pa(pud) | _KERNPG_TABLE));
152 pud = pud_offset(p4d, vaddr);
153 if (!pud_present(*pud)) {
154 pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
157 image->arch.pmd = pmd;
158 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
160 pmd = pmd_offset(pud, vaddr);
161 if (!pmd_present(*pmd)) {
162 pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
165 image->arch.pte = pte;
166 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
168 pte = pte_offset_kernel(pmd, vaddr);
171 prot = PAGE_KERNEL_EXEC;
173 set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, prot));
179 static void *alloc_pgt_page(void *data)
181 struct kimage *image = (struct kimage *)data;
185 page = kimage_alloc_control_pages(image, 0);
187 p = page_address(page);
194 static int init_pgtable(struct kimage *image, unsigned long start_pgtable)
196 struct x86_mapping_info info = {
197 .alloc_pgt_page = alloc_pgt_page,
199 .page_flag = __PAGE_KERNEL_LARGE_EXEC,
200 .kernpg_flag = _KERNPG_TABLE_NOENC,
202 unsigned long mstart, mend;
207 level4p = (pgd_t *)__va(start_pgtable);
211 info.page_flag |= _PAGE_ENC;
212 info.kernpg_flag |= _PAGE_ENC;
216 info.direct_gbpages = true;
218 for (i = 0; i < nr_pfn_mapped; i++) {
219 mstart = pfn_mapped[i].start << PAGE_SHIFT;
220 mend = pfn_mapped[i].end << PAGE_SHIFT;
222 result = kernel_ident_mapping_init(&info,
223 level4p, mstart, mend);
229 * segments's mem ranges could be outside 0 ~ max_pfn,
230 * for example when jump back to original kernel from kexeced kernel.
231 * or first kernel is booted with user mem map, and second kernel
232 * could be loaded out of that range.
234 for (i = 0; i < image->nr_segments; i++) {
235 mstart = image->segment[i].mem;
236 mend = mstart + image->segment[i].memsz;
238 result = kernel_ident_mapping_init(&info,
239 level4p, mstart, mend);
246 * Prepare EFI systab and ACPI tables for kexec kernel since they are
247 * not covered by pfn_mapped.
249 result = map_efi_systab(&info, level4p);
253 result = map_acpi_tables(&info, level4p);
257 return init_transition_pgtable(image, level4p);
260 static void set_idt(void *newidt, u16 limit)
262 struct desc_ptr curidt;
264 /* x86-64 supports unaliged loads & stores */
266 curidt.address = (unsigned long)newidt;
268 __asm__ __volatile__ (
275 static void set_gdt(void *newgdt, u16 limit)
277 struct desc_ptr curgdt;
279 /* x86-64 supports unaligned loads & stores */
281 curgdt.address = (unsigned long)newgdt;
283 __asm__ __volatile__ (
289 static void load_segments(void)
291 __asm__ __volatile__ (
297 : : "a" (__KERNEL_DS) : "memory"
301 #ifdef CONFIG_KEXEC_FILE
302 /* Update purgatory as needed after various image segments have been prepared */
303 static int arch_update_purgatory(struct kimage *image)
307 if (!image->file_mode)
310 /* Setup copying of backup region */
311 if (image->type == KEXEC_TYPE_CRASH) {
312 ret = kexec_purgatory_get_set_symbol(image,
313 "purgatory_backup_dest",
314 &image->arch.backup_load_addr,
315 sizeof(image->arch.backup_load_addr), 0);
319 ret = kexec_purgatory_get_set_symbol(image,
320 "purgatory_backup_src",
321 &image->arch.backup_src_start,
322 sizeof(image->arch.backup_src_start), 0);
326 ret = kexec_purgatory_get_set_symbol(image,
327 "purgatory_backup_sz",
328 &image->arch.backup_src_sz,
329 sizeof(image->arch.backup_src_sz), 0);
336 #else /* !CONFIG_KEXEC_FILE */
337 static inline int arch_update_purgatory(struct kimage *image)
341 #endif /* CONFIG_KEXEC_FILE */
343 int machine_kexec_prepare(struct kimage *image)
345 unsigned long start_pgtable;
348 /* Calculate the offsets */
349 start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
351 /* Setup the identity mapped 64bit page table */
352 result = init_pgtable(image, start_pgtable);
356 /* update purgatory as needed */
357 result = arch_update_purgatory(image);
364 void machine_kexec_cleanup(struct kimage *image)
366 free_transition_pgtable(image);
370 * Do not allocate memory (or fail in any way) in machine_kexec().
371 * We are past the point of no return, committed to rebooting now.
373 void machine_kexec(struct kimage *image)
375 unsigned long page_list[PAGES_NR];
377 int save_ftrace_enabled;
379 #ifdef CONFIG_KEXEC_JUMP
380 if (image->preserve_context)
381 save_processor_state();
384 save_ftrace_enabled = __ftrace_enabled_save();
386 /* Interrupts aren't acceptable while we reboot */
388 hw_breakpoint_disable();
390 if (image->preserve_context) {
391 #ifdef CONFIG_X86_IO_APIC
393 * We need to put APICs in legacy mode so that we can
394 * get timer interrupts in second kernel. kexec/kdump
395 * paths already have calls to restore_boot_irq_mode()
396 * in one form or other. kexec jump path also need one.
399 restore_boot_irq_mode();
403 control_page = page_address(image->control_code_page) + PAGE_SIZE;
404 memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
406 page_list[PA_CONTROL_PAGE] = virt_to_phys(control_page);
407 page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
408 page_list[PA_TABLE_PAGE] =
409 (unsigned long)__pa(page_address(image->control_code_page));
411 if (image->type == KEXEC_TYPE_DEFAULT)
412 page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
416 * The segment registers are funny things, they have both a
417 * visible and an invisible part. Whenever the visible part is
418 * set to a specific selector, the invisible part is loaded
419 * with from a table in memory. At no other time is the
420 * descriptor table in memory accessed.
422 * I take advantage of this here by force loading the
423 * segments, before I zap the gdt with an invalid value.
427 * The gdt & idt are now invalid.
428 * If you want to load them you must set up your own idt & gdt.
430 set_gdt(phys_to_virt(0), 0);
431 set_idt(phys_to_virt(0), 0);
434 image->start = relocate_kernel((unsigned long)image->head,
435 (unsigned long)page_list,
437 image->preserve_context,
440 #ifdef CONFIG_KEXEC_JUMP
441 if (image->preserve_context)
442 restore_processor_state();
445 __ftrace_enabled_restore(save_ftrace_enabled);
448 void arch_crash_save_vmcoreinfo(void)
450 u64 sme_mask = sme_me_mask;
452 VMCOREINFO_NUMBER(phys_base);
453 VMCOREINFO_SYMBOL(init_top_pgt);
454 vmcoreinfo_append_str("NUMBER(pgtable_l5_enabled)=%d\n",
455 pgtable_l5_enabled());
458 VMCOREINFO_SYMBOL(node_data);
459 VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);
461 vmcoreinfo_append_str("KERNELOFFSET=%lx\n",
463 VMCOREINFO_NUMBER(KERNEL_IMAGE_SIZE);
464 VMCOREINFO_NUMBER(sme_mask);
467 /* arch-dependent functionality related to kexec file-based syscall */
469 #ifdef CONFIG_KEXEC_FILE
470 void *arch_kexec_kernel_image_load(struct kimage *image)
472 vfree(image->arch.elf_headers);
473 image->arch.elf_headers = NULL;
475 if (!image->fops || !image->fops->load)
476 return ERR_PTR(-ENOEXEC);
478 return image->fops->load(image, image->kernel_buf,
479 image->kernel_buf_len, image->initrd_buf,
480 image->initrd_buf_len, image->cmdline_buf,
481 image->cmdline_buf_len);
485 * Apply purgatory relocations.
487 * @pi: Purgatory to be relocated.
488 * @section: Section relocations applying to.
489 * @relsec: Section containing RELAs.
490 * @symtabsec: Corresponding symtab.
492 * TODO: Some of the code belongs to generic code. Move that in kexec.c.
494 int arch_kexec_apply_relocations_add(struct purgatory_info *pi,
495 Elf_Shdr *section, const Elf_Shdr *relsec,
496 const Elf_Shdr *symtabsec)
502 unsigned long address, sec_base, value;
503 const char *strtab, *name, *shstrtab;
504 const Elf_Shdr *sechdrs;
506 /* String & section header string table */
507 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
508 strtab = (char *)pi->ehdr + sechdrs[symtabsec->sh_link].sh_offset;
509 shstrtab = (char *)pi->ehdr + sechdrs[pi->ehdr->e_shstrndx].sh_offset;
511 rel = (void *)pi->ehdr + relsec->sh_offset;
513 pr_debug("Applying relocate section %s to %u\n",
514 shstrtab + relsec->sh_name, relsec->sh_info);
516 for (i = 0; i < relsec->sh_size / sizeof(*rel); i++) {
519 * rel[i].r_offset contains byte offset from beginning
520 * of section to the storage unit affected.
522 * This is location to update. This is temporary buffer
523 * where section is currently loaded. This will finally be
524 * loaded to a different address later, pointed to by
525 * ->sh_addr. kexec takes care of moving it
526 * (kexec_load_segment()).
528 location = pi->purgatory_buf;
529 location += section->sh_offset;
530 location += rel[i].r_offset;
532 /* Final address of the location */
533 address = section->sh_addr + rel[i].r_offset;
536 * rel[i].r_info contains information about symbol table index
537 * w.r.t which relocation must be made and type of relocation
538 * to apply. ELF64_R_SYM() and ELF64_R_TYPE() macros get
539 * these respectively.
541 sym = (void *)pi->ehdr + symtabsec->sh_offset;
542 sym += ELF64_R_SYM(rel[i].r_info);
545 name = strtab + sym->st_name;
547 name = shstrtab + sechdrs[sym->st_shndx].sh_name;
549 pr_debug("Symbol: %s info: %02x shndx: %02x value=%llx size: %llx\n",
550 name, sym->st_info, sym->st_shndx, sym->st_value,
553 if (sym->st_shndx == SHN_UNDEF) {
554 pr_err("Undefined symbol: %s\n", name);
558 if (sym->st_shndx == SHN_COMMON) {
559 pr_err("symbol '%s' in common section\n", name);
563 if (sym->st_shndx == SHN_ABS)
565 else if (sym->st_shndx >= pi->ehdr->e_shnum) {
566 pr_err("Invalid section %d for symbol %s\n",
567 sym->st_shndx, name);
570 sec_base = pi->sechdrs[sym->st_shndx].sh_addr;
572 value = sym->st_value;
574 value += rel[i].r_addend;
576 switch (ELF64_R_TYPE(rel[i].r_info)) {
580 *(u64 *)location = value;
583 *(u32 *)location = value;
584 if (value != *(u32 *)location)
588 *(s32 *)location = value;
589 if ((s64)value != *(s32 *)location)
594 value -= (u64)address;
595 *(u32 *)location = value;
598 pr_err("Unknown rela relocation: %llu\n",
599 ELF64_R_TYPE(rel[i].r_info));
606 pr_err("Overflow in relocation type %d value 0x%lx\n",
607 (int)ELF64_R_TYPE(rel[i].r_info), value);
610 #endif /* CONFIG_KEXEC_FILE */
613 kexec_mark_range(unsigned long start, unsigned long end, bool protect)
616 unsigned int nr_pages;
619 * For physical range: [start, end]. We must skip the unassigned
620 * crashk resource with zero-valued "end" member.
622 if (!end || start > end)
625 page = pfn_to_page(start >> PAGE_SHIFT);
626 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
628 return set_pages_ro(page, nr_pages);
630 return set_pages_rw(page, nr_pages);
633 static void kexec_mark_crashkres(bool protect)
635 unsigned long control;
637 kexec_mark_range(crashk_low_res.start, crashk_low_res.end, protect);
639 /* Don't touch the control code page used in crash_kexec().*/
640 control = PFN_PHYS(page_to_pfn(kexec_crash_image->control_code_page));
641 /* Control code page is located in the 2nd page. */
642 kexec_mark_range(crashk_res.start, control + PAGE_SIZE - 1, protect);
643 control += KEXEC_CONTROL_PAGE_SIZE;
644 kexec_mark_range(control, crashk_res.end, protect);
647 void arch_kexec_protect_crashkres(void)
649 kexec_mark_crashkres(true);
652 void arch_kexec_unprotect_crashkres(void)
654 kexec_mark_crashkres(false);
658 * During a traditional boot under SME, SME will encrypt the kernel,
659 * so the SME kexec kernel also needs to be un-encrypted in order to
660 * replicate a normal SME boot.
662 * During a traditional boot under SEV, the kernel has already been
663 * loaded encrypted, so the SEV kexec kernel needs to be encrypted in
664 * order to replicate a normal SEV boot.
666 int arch_kexec_post_alloc_pages(void *vaddr, unsigned int pages, gfp_t gfp)
672 * If SME is active we need to be sure that kexec pages are
673 * not encrypted because when we boot to the new kernel the
674 * pages won't be accessed encrypted (initially).
676 return set_memory_decrypted((unsigned long)vaddr, pages);
679 void arch_kexec_pre_free_pages(void *vaddr, unsigned int pages)
685 * If SME is active we need to reset the pages back to being
686 * an encrypted mapping before freeing them.
688 set_memory_encrypted((unsigned long)vaddr, pages);