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/tlbflush.h>
23 #include <asm/mmu_context.h>
24 #include <asm/io_apic.h>
25 #include <asm/debugreg.h>
26 #include <asm/kexec-bzimage64.h>
27 #include <asm/setup.h>
28 #include <asm/set_memory.h>
32 * Used while adding mapping for ACPI tables.
33 * Can be reused when other iomem regions need be mapped
35 struct init_pgtable_data {
36 struct x86_mapping_info *info;
40 static int mem_region_callback(struct resource *res, void *arg)
42 struct init_pgtable_data *data = arg;
43 unsigned long mstart, mend;
46 mend = mstart + resource_size(res) - 1;
48 return kernel_ident_mapping_init(data->info, data->level4p, mstart, mend);
52 map_acpi_tables(struct x86_mapping_info *info, pgd_t *level4p)
54 struct init_pgtable_data data;
59 data.level4p = level4p;
60 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
62 ret = walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1,
63 &data, mem_region_callback);
64 if (ret && ret != -EINVAL)
67 /* ACPI tables could be located in ACPI Non-volatile Storage region */
68 ret = walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1,
69 &data, mem_region_callback);
70 if (ret && ret != -EINVAL)
76 static int map_acpi_tables(struct x86_mapping_info *info, pgd_t *level4p) { return 0; }
79 #ifdef CONFIG_KEXEC_FILE
80 const struct kexec_file_ops * const kexec_file_loaders[] = {
87 map_efi_systab(struct x86_mapping_info *info, pgd_t *level4p)
90 unsigned long mstart, mend;
92 if (!efi_enabled(EFI_BOOT))
95 mstart = (boot_params.efi_info.efi_systab |
96 ((u64)boot_params.efi_info.efi_systab_hi<<32));
98 if (efi_enabled(EFI_64BIT))
99 mend = mstart + sizeof(efi_system_table_64_t);
101 mend = mstart + sizeof(efi_system_table_32_t);
106 return kernel_ident_mapping_init(info, level4p, mstart, mend);
111 static void free_transition_pgtable(struct kimage *image)
113 free_page((unsigned long)image->arch.p4d);
114 image->arch.p4d = NULL;
115 free_page((unsigned long)image->arch.pud);
116 image->arch.pud = NULL;
117 free_page((unsigned long)image->arch.pmd);
118 image->arch.pmd = NULL;
119 free_page((unsigned long)image->arch.pte);
120 image->arch.pte = NULL;
123 static int init_transition_pgtable(struct kimage *image, pgd_t *pgd)
125 pgprot_t prot = PAGE_KERNEL_EXEC_NOENC;
126 unsigned long vaddr, paddr;
127 int result = -ENOMEM;
133 vaddr = (unsigned long)relocate_kernel;
134 paddr = __pa(page_address(image->control_code_page)+PAGE_SIZE);
135 pgd += pgd_index(vaddr);
136 if (!pgd_present(*pgd)) {
137 p4d = (p4d_t *)get_zeroed_page(GFP_KERNEL);
140 image->arch.p4d = p4d;
141 set_pgd(pgd, __pgd(__pa(p4d) | _KERNPG_TABLE));
143 p4d = p4d_offset(pgd, vaddr);
144 if (!p4d_present(*p4d)) {
145 pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
148 image->arch.pud = pud;
149 set_p4d(p4d, __p4d(__pa(pud) | _KERNPG_TABLE));
151 pud = pud_offset(p4d, vaddr);
152 if (!pud_present(*pud)) {
153 pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
156 image->arch.pmd = pmd;
157 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
159 pmd = pmd_offset(pud, vaddr);
160 if (!pmd_present(*pmd)) {
161 pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
164 image->arch.pte = pte;
165 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
167 pte = pte_offset_kernel(pmd, vaddr);
170 prot = PAGE_KERNEL_EXEC;
172 set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, prot));
178 static void *alloc_pgt_page(void *data)
180 struct kimage *image = (struct kimage *)data;
184 page = kimage_alloc_control_pages(image, 0);
186 p = page_address(page);
193 static int init_pgtable(struct kimage *image, unsigned long start_pgtable)
195 struct x86_mapping_info info = {
196 .alloc_pgt_page = alloc_pgt_page,
198 .page_flag = __PAGE_KERNEL_LARGE_EXEC,
199 .kernpg_flag = _KERNPG_TABLE_NOENC,
201 unsigned long mstart, mend;
206 level4p = (pgd_t *)__va(start_pgtable);
210 info.page_flag |= _PAGE_ENC;
211 info.kernpg_flag |= _PAGE_ENC;
215 info.direct_gbpages = true;
217 for (i = 0; i < nr_pfn_mapped; i++) {
218 mstart = pfn_mapped[i].start << PAGE_SHIFT;
219 mend = pfn_mapped[i].end << PAGE_SHIFT;
221 result = kernel_ident_mapping_init(&info,
222 level4p, mstart, mend);
228 * segments's mem ranges could be outside 0 ~ max_pfn,
229 * for example when jump back to original kernel from kexeced kernel.
230 * or first kernel is booted with user mem map, and second kernel
231 * could be loaded out of that range.
233 for (i = 0; i < image->nr_segments; i++) {
234 mstart = image->segment[i].mem;
235 mend = mstart + image->segment[i].memsz;
237 result = kernel_ident_mapping_init(&info,
238 level4p, mstart, mend);
245 * Prepare EFI systab and ACPI tables for kexec kernel since they are
246 * not covered by pfn_mapped.
248 result = map_efi_systab(&info, level4p);
252 result = map_acpi_tables(&info, level4p);
256 return init_transition_pgtable(image, level4p);
259 static void load_segments(void)
261 __asm__ __volatile__ (
267 : : "a" (__KERNEL_DS) : "memory"
271 int machine_kexec_prepare(struct kimage *image)
273 unsigned long start_pgtable;
276 /* Calculate the offsets */
277 start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
279 /* Setup the identity mapped 64bit page table */
280 result = init_pgtable(image, start_pgtable);
287 void machine_kexec_cleanup(struct kimage *image)
289 free_transition_pgtable(image);
293 * Do not allocate memory (or fail in any way) in machine_kexec().
294 * We are past the point of no return, committed to rebooting now.
296 void machine_kexec(struct kimage *image)
298 unsigned long page_list[PAGES_NR];
300 int save_ftrace_enabled;
302 #ifdef CONFIG_KEXEC_JUMP
303 if (image->preserve_context)
304 save_processor_state();
307 save_ftrace_enabled = __ftrace_enabled_save();
309 /* Interrupts aren't acceptable while we reboot */
311 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,
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
374 void *arch_kexec_kernel_image_load(struct kimage *image)
376 if (!image->fops || !image->fops->load)
377 return ERR_PTR(-ENOEXEC);
379 return image->fops->load(image, image->kernel_buf,
380 image->kernel_buf_len, image->initrd_buf,
381 image->initrd_buf_len, image->cmdline_buf,
382 image->cmdline_buf_len);
386 * Apply purgatory relocations.
388 * @pi: Purgatory to be relocated.
389 * @section: Section relocations applying to.
390 * @relsec: Section containing RELAs.
391 * @symtabsec: Corresponding symtab.
393 * TODO: Some of the code belongs to generic code. Move that in kexec.c.
395 int arch_kexec_apply_relocations_add(struct purgatory_info *pi,
396 Elf_Shdr *section, const Elf_Shdr *relsec,
397 const Elf_Shdr *symtabsec)
403 unsigned long address, sec_base, value;
404 const char *strtab, *name, *shstrtab;
405 const Elf_Shdr *sechdrs;
407 /* String & section header string table */
408 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
409 strtab = (char *)pi->ehdr + sechdrs[symtabsec->sh_link].sh_offset;
410 shstrtab = (char *)pi->ehdr + sechdrs[pi->ehdr->e_shstrndx].sh_offset;
412 rel = (void *)pi->ehdr + relsec->sh_offset;
414 pr_debug("Applying relocate section %s to %u\n",
415 shstrtab + relsec->sh_name, relsec->sh_info);
417 for (i = 0; i < relsec->sh_size / sizeof(*rel); i++) {
420 * rel[i].r_offset contains byte offset from beginning
421 * of section to the storage unit affected.
423 * This is location to update. This is temporary buffer
424 * where section is currently loaded. This will finally be
425 * loaded to a different address later, pointed to by
426 * ->sh_addr. kexec takes care of moving it
427 * (kexec_load_segment()).
429 location = pi->purgatory_buf;
430 location += section->sh_offset;
431 location += rel[i].r_offset;
433 /* Final address of the location */
434 address = section->sh_addr + rel[i].r_offset;
437 * rel[i].r_info contains information about symbol table index
438 * w.r.t which relocation must be made and type of relocation
439 * to apply. ELF64_R_SYM() and ELF64_R_TYPE() macros get
440 * these respectively.
442 sym = (void *)pi->ehdr + symtabsec->sh_offset;
443 sym += ELF64_R_SYM(rel[i].r_info);
446 name = strtab + sym->st_name;
448 name = shstrtab + sechdrs[sym->st_shndx].sh_name;
450 pr_debug("Symbol: %s info: %02x shndx: %02x value=%llx size: %llx\n",
451 name, sym->st_info, sym->st_shndx, sym->st_value,
454 if (sym->st_shndx == SHN_UNDEF) {
455 pr_err("Undefined symbol: %s\n", name);
459 if (sym->st_shndx == SHN_COMMON) {
460 pr_err("symbol '%s' in common section\n", name);
464 if (sym->st_shndx == SHN_ABS)
466 else if (sym->st_shndx >= pi->ehdr->e_shnum) {
467 pr_err("Invalid section %d for symbol %s\n",
468 sym->st_shndx, name);
471 sec_base = pi->sechdrs[sym->st_shndx].sh_addr;
473 value = sym->st_value;
475 value += rel[i].r_addend;
477 switch (ELF64_R_TYPE(rel[i].r_info)) {
481 *(u64 *)location = value;
484 *(u32 *)location = value;
485 if (value != *(u32 *)location)
489 *(s32 *)location = value;
490 if ((s64)value != *(s32 *)location)
495 value -= (u64)address;
496 *(u32 *)location = value;
499 pr_err("Unknown rela relocation: %llu\n",
500 ELF64_R_TYPE(rel[i].r_info));
507 pr_err("Overflow in relocation type %d value 0x%lx\n",
508 (int)ELF64_R_TYPE(rel[i].r_info), value);
512 int arch_kimage_file_post_load_cleanup(struct kimage *image)
514 vfree(image->elf_headers);
515 image->elf_headers = NULL;
516 image->elf_headers_sz = 0;
518 return kexec_image_post_load_cleanup_default(image);
520 #endif /* CONFIG_KEXEC_FILE */
523 kexec_mark_range(unsigned long start, unsigned long end, bool protect)
526 unsigned int nr_pages;
529 * For physical range: [start, end]. We must skip the unassigned
530 * crashk resource with zero-valued "end" member.
532 if (!end || start > end)
535 page = pfn_to_page(start >> PAGE_SHIFT);
536 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
538 return set_pages_ro(page, nr_pages);
540 return set_pages_rw(page, nr_pages);
543 static void kexec_mark_crashkres(bool protect)
545 unsigned long control;
547 kexec_mark_range(crashk_low_res.start, crashk_low_res.end, protect);
549 /* Don't touch the control code page used in crash_kexec().*/
550 control = PFN_PHYS(page_to_pfn(kexec_crash_image->control_code_page));
551 /* Control code page is located in the 2nd page. */
552 kexec_mark_range(crashk_res.start, control + PAGE_SIZE - 1, protect);
553 control += KEXEC_CONTROL_PAGE_SIZE;
554 kexec_mark_range(control, crashk_res.end, protect);
557 void arch_kexec_protect_crashkres(void)
559 kexec_mark_crashkres(true);
562 void arch_kexec_unprotect_crashkres(void)
564 kexec_mark_crashkres(false);
568 * During a traditional boot under SME, SME will encrypt the kernel,
569 * so the SME kexec kernel also needs to be un-encrypted in order to
570 * replicate a normal SME boot.
572 * During a traditional boot under SEV, the kernel has already been
573 * loaded encrypted, so the SEV kexec kernel needs to be encrypted in
574 * order to replicate a normal SEV boot.
576 int arch_kexec_post_alloc_pages(void *vaddr, unsigned int pages, gfp_t gfp)
582 * If SME is active we need to be sure that kexec pages are
583 * not encrypted because when we boot to the new kernel the
584 * pages won't be accessed encrypted (initially).
586 return set_memory_decrypted((unsigned long)vaddr, pages);
589 void arch_kexec_pre_free_pages(void *vaddr, unsigned int pages)
595 * If SME is active we need to reset the pages back to being
596 * an encrypted mapping before freeing them.
598 set_memory_encrypted((unsigned long)vaddr, pages);