2 * handle transition of Linux booting another kernel
3 * Copyright (C) 2002-2005 Eric Biederman <ebiederm@xmission.com>
5 * This source code is licensed under the GNU General Public License,
6 * Version 2. See the file COPYING for more details.
9 #define pr_fmt(fmt) "kexec: " fmt
12 #include <linux/kexec.h>
13 #include <linux/string.h>
14 #include <linux/gfp.h>
15 #include <linux/reboot.h>
16 #include <linux/numa.h>
17 #include <linux/ftrace.h>
19 #include <linux/suspend.h>
20 #include <linux/vmalloc.h>
23 #include <asm/pgtable.h>
24 #include <asm/tlbflush.h>
25 #include <asm/mmu_context.h>
26 #include <asm/io_apic.h>
27 #include <asm/debugreg.h>
28 #include <asm/kexec-bzimage64.h>
29 #include <asm/setup.h>
31 #ifdef CONFIG_KEXEC_FILE
32 static struct kexec_file_ops *kexec_file_loaders[] = {
37 static void free_transition_pgtable(struct kimage *image)
39 free_page((unsigned long)image->arch.pud);
40 free_page((unsigned long)image->arch.pmd);
41 free_page((unsigned long)image->arch.pte);
44 static int init_transition_pgtable(struct kimage *image, pgd_t *pgd)
49 unsigned long vaddr, paddr;
52 vaddr = (unsigned long)relocate_kernel;
53 paddr = __pa(page_address(image->control_code_page)+PAGE_SIZE);
54 pgd += pgd_index(vaddr);
55 if (!pgd_present(*pgd)) {
56 pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
59 image->arch.pud = pud;
60 set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE));
62 pud = pud_offset(pgd, vaddr);
63 if (!pud_present(*pud)) {
64 pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
67 image->arch.pmd = pmd;
68 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
70 pmd = pmd_offset(pud, vaddr);
71 if (!pmd_present(*pmd)) {
72 pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
75 image->arch.pte = pte;
76 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
78 pte = pte_offset_kernel(pmd, vaddr);
79 set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC));
82 free_transition_pgtable(image);
86 static void *alloc_pgt_page(void *data)
88 struct kimage *image = (struct kimage *)data;
92 page = kimage_alloc_control_pages(image, 0);
94 p = page_address(page);
101 static int init_pgtable(struct kimage *image, unsigned long start_pgtable)
103 struct x86_mapping_info info = {
104 .alloc_pgt_page = alloc_pgt_page,
106 .pmd_flag = __PAGE_KERNEL_LARGE_EXEC,
108 unsigned long mstart, mend;
113 level4p = (pgd_t *)__va(start_pgtable);
115 for (i = 0; i < nr_pfn_mapped; i++) {
116 mstart = pfn_mapped[i].start << PAGE_SHIFT;
117 mend = pfn_mapped[i].end << PAGE_SHIFT;
119 result = kernel_ident_mapping_init(&info,
120 level4p, mstart, mend);
126 * segments's mem ranges could be outside 0 ~ max_pfn,
127 * for example when jump back to original kernel from kexeced kernel.
128 * or first kernel is booted with user mem map, and second kernel
129 * could be loaded out of that range.
131 for (i = 0; i < image->nr_segments; i++) {
132 mstart = image->segment[i].mem;
133 mend = mstart + image->segment[i].memsz;
135 result = kernel_ident_mapping_init(&info,
136 level4p, mstart, mend);
142 return init_transition_pgtable(image, level4p);
145 static void set_idt(void *newidt, u16 limit)
147 struct desc_ptr curidt;
149 /* x86-64 supports unaliged loads & stores */
151 curidt.address = (unsigned long)newidt;
153 __asm__ __volatile__ (
160 static void set_gdt(void *newgdt, u16 limit)
162 struct desc_ptr curgdt;
164 /* x86-64 supports unaligned loads & stores */
166 curgdt.address = (unsigned long)newgdt;
168 __asm__ __volatile__ (
174 static void load_segments(void)
176 __asm__ __volatile__ (
182 : : "a" (__KERNEL_DS) : "memory"
186 #ifdef CONFIG_KEXEC_FILE
187 /* Update purgatory as needed after various image segments have been prepared */
188 static int arch_update_purgatory(struct kimage *image)
192 if (!image->file_mode)
195 /* Setup copying of backup region */
196 if (image->type == KEXEC_TYPE_CRASH) {
197 ret = kexec_purgatory_get_set_symbol(image,
198 "purgatory_backup_dest",
199 &image->arch.backup_load_addr,
200 sizeof(image->arch.backup_load_addr), 0);
204 ret = kexec_purgatory_get_set_symbol(image,
205 "purgatory_backup_src",
206 &image->arch.backup_src_start,
207 sizeof(image->arch.backup_src_start), 0);
211 ret = kexec_purgatory_get_set_symbol(image,
212 "purgatory_backup_sz",
213 &image->arch.backup_src_sz,
214 sizeof(image->arch.backup_src_sz), 0);
221 #else /* !CONFIG_KEXEC_FILE */
222 static inline int arch_update_purgatory(struct kimage *image)
226 #endif /* CONFIG_KEXEC_FILE */
228 int machine_kexec_prepare(struct kimage *image)
230 unsigned long start_pgtable;
233 /* Calculate the offsets */
234 start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
236 /* Setup the identity mapped 64bit page table */
237 result = init_pgtable(image, start_pgtable);
241 /* update purgatory as needed */
242 result = arch_update_purgatory(image);
249 void machine_kexec_cleanup(struct kimage *image)
251 free_transition_pgtable(image);
255 * Do not allocate memory (or fail in any way) in machine_kexec().
256 * We are past the point of no return, committed to rebooting now.
258 void machine_kexec(struct kimage *image)
260 unsigned long page_list[PAGES_NR];
262 int save_ftrace_enabled;
264 #ifdef CONFIG_KEXEC_JUMP
265 if (image->preserve_context)
266 save_processor_state();
269 save_ftrace_enabled = __ftrace_enabled_save();
271 /* Interrupts aren't acceptable while we reboot */
273 hw_breakpoint_disable();
275 if (image->preserve_context) {
276 #ifdef CONFIG_X86_IO_APIC
278 * We need to put APICs in legacy mode so that we can
279 * get timer interrupts in second kernel. kexec/kdump
280 * paths already have calls to disable_IO_APIC() in
281 * one form or other. kexec jump path also need
288 control_page = page_address(image->control_code_page) + PAGE_SIZE;
289 memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
291 page_list[PA_CONTROL_PAGE] = virt_to_phys(control_page);
292 page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
293 page_list[PA_TABLE_PAGE] =
294 (unsigned long)__pa(page_address(image->control_code_page));
296 if (image->type == KEXEC_TYPE_DEFAULT)
297 page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
301 * The segment registers are funny things, they have both a
302 * visible and an invisible part. Whenever the visible part is
303 * set to a specific selector, the invisible part is loaded
304 * with from a table in memory. At no other time is the
305 * descriptor table in memory accessed.
307 * I take advantage of this here by force loading the
308 * segments, before I zap the gdt with an invalid value.
312 * The gdt & idt are now invalid.
313 * If you want to load them you must set up your own idt & gdt.
315 set_gdt(phys_to_virt(0), 0);
316 set_idt(phys_to_virt(0), 0);
319 image->start = relocate_kernel((unsigned long)image->head,
320 (unsigned long)page_list,
322 image->preserve_context);
324 #ifdef CONFIG_KEXEC_JUMP
325 if (image->preserve_context)
326 restore_processor_state();
329 __ftrace_enabled_restore(save_ftrace_enabled);
332 void arch_crash_save_vmcoreinfo(void)
334 VMCOREINFO_NUMBER(phys_base);
335 VMCOREINFO_SYMBOL(init_level4_pgt);
338 VMCOREINFO_SYMBOL(node_data);
339 VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);
341 vmcoreinfo_append_str("KERNELOFFSET=%lx\n",
343 VMCOREINFO_NUMBER(KERNEL_IMAGE_SIZE);
346 /* arch-dependent functionality related to kexec file-based syscall */
348 #ifdef CONFIG_KEXEC_FILE
349 int arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
350 unsigned long buf_len)
352 int i, ret = -ENOEXEC;
353 struct kexec_file_ops *fops;
355 for (i = 0; i < ARRAY_SIZE(kexec_file_loaders); i++) {
356 fops = kexec_file_loaders[i];
357 if (!fops || !fops->probe)
360 ret = fops->probe(buf, buf_len);
370 void *arch_kexec_kernel_image_load(struct kimage *image)
372 vfree(image->arch.elf_headers);
373 image->arch.elf_headers = NULL;
375 if (!image->fops || !image->fops->load)
376 return ERR_PTR(-ENOEXEC);
378 return image->fops->load(image, image->kernel_buf,
379 image->kernel_buf_len, image->initrd_buf,
380 image->initrd_buf_len, image->cmdline_buf,
381 image->cmdline_buf_len);
384 int arch_kimage_file_post_load_cleanup(struct kimage *image)
386 if (!image->fops || !image->fops->cleanup)
389 return image->fops->cleanup(image->image_loader_data);
392 #ifdef CONFIG_KEXEC_VERIFY_SIG
393 int arch_kexec_kernel_verify_sig(struct kimage *image, void *kernel,
394 unsigned long kernel_len)
396 if (!image->fops || !image->fops->verify_sig) {
397 pr_debug("kernel loader does not support signature verification.");
398 return -EKEYREJECTED;
401 return image->fops->verify_sig(kernel, kernel_len);
406 * Apply purgatory relocations.
408 * ehdr: Pointer to elf headers
409 * sechdrs: Pointer to section headers.
410 * relsec: section index of SHT_RELA section.
412 * TODO: Some of the code belongs to generic code. Move that in kexec.c.
414 int arch_kexec_apply_relocations_add(const Elf64_Ehdr *ehdr,
415 Elf64_Shdr *sechdrs, unsigned int relsec)
421 Elf64_Shdr *section, *symtabsec;
422 unsigned long address, sec_base, value;
423 const char *strtab, *name, *shstrtab;
426 * ->sh_offset has been modified to keep the pointer to section
429 rel = (void *)sechdrs[relsec].sh_offset;
431 /* Section to which relocations apply */
432 section = &sechdrs[sechdrs[relsec].sh_info];
434 pr_debug("Applying relocate section %u to %u\n", relsec,
435 sechdrs[relsec].sh_info);
437 /* Associated symbol table */
438 symtabsec = &sechdrs[sechdrs[relsec].sh_link];
441 if (symtabsec->sh_link >= ehdr->e_shnum) {
442 /* Invalid strtab section number */
443 pr_err("Invalid string table section index %d\n",
448 strtab = (char *)sechdrs[symtabsec->sh_link].sh_offset;
450 /* section header string table */
451 shstrtab = (char *)sechdrs[ehdr->e_shstrndx].sh_offset;
453 for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rel); i++) {
456 * rel[i].r_offset contains byte offset from beginning
457 * of section to the storage unit affected.
459 * This is location to update (->sh_offset). This is temporary
460 * buffer where section is currently loaded. This will finally
461 * be loaded to a different address later, pointed to by
462 * ->sh_addr. kexec takes care of moving it
463 * (kexec_load_segment()).
465 location = (void *)(section->sh_offset + rel[i].r_offset);
467 /* Final address of the location */
468 address = section->sh_addr + rel[i].r_offset;
471 * rel[i].r_info contains information about symbol table index
472 * w.r.t which relocation must be made and type of relocation
473 * to apply. ELF64_R_SYM() and ELF64_R_TYPE() macros get
474 * these respectively.
476 sym = (Elf64_Sym *)symtabsec->sh_offset +
477 ELF64_R_SYM(rel[i].r_info);
480 name = strtab + sym->st_name;
482 name = shstrtab + sechdrs[sym->st_shndx].sh_name;
484 pr_debug("Symbol: %s info: %02x shndx: %02x value=%llx size: %llx\n",
485 name, sym->st_info, sym->st_shndx, sym->st_value,
488 if (sym->st_shndx == SHN_UNDEF) {
489 pr_err("Undefined symbol: %s\n", name);
493 if (sym->st_shndx == SHN_COMMON) {
494 pr_err("symbol '%s' in common section\n", name);
498 if (sym->st_shndx == SHN_ABS)
500 else if (sym->st_shndx >= ehdr->e_shnum) {
501 pr_err("Invalid section %d for symbol %s\n",
502 sym->st_shndx, name);
505 sec_base = sechdrs[sym->st_shndx].sh_addr;
507 value = sym->st_value;
509 value += rel[i].r_addend;
511 switch (ELF64_R_TYPE(rel[i].r_info)) {
515 *(u64 *)location = value;
518 *(u32 *)location = value;
519 if (value != *(u32 *)location)
523 *(s32 *)location = value;
524 if ((s64)value != *(s32 *)location)
528 value -= (u64)address;
529 *(u32 *)location = value;
532 pr_err("Unknown rela relocation: %llu\n",
533 ELF64_R_TYPE(rel[i].r_info));
540 pr_err("Overflow in relocation type %d value 0x%lx\n",
541 (int)ELF64_R_TYPE(rel[i].r_info), value);
544 #endif /* CONFIG_KEXEC_FILE */
547 kexec_mark_range(unsigned long start, unsigned long end, bool protect)
550 unsigned int nr_pages;
553 * For physical range: [start, end]. We must skip the unassigned
554 * crashk resource with zero-valued "end" member.
556 if (!end || start > end)
559 page = pfn_to_page(start >> PAGE_SHIFT);
560 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
562 return set_pages_ro(page, nr_pages);
564 return set_pages_rw(page, nr_pages);
567 static void kexec_mark_crashkres(bool protect)
569 unsigned long control;
571 kexec_mark_range(crashk_low_res.start, crashk_low_res.end, protect);
573 /* Don't touch the control code page used in crash_kexec().*/
574 control = PFN_PHYS(page_to_pfn(kexec_crash_image->control_code_page));
575 /* Control code page is located in the 2nd page. */
576 kexec_mark_range(crashk_res.start, control + PAGE_SIZE - 1, protect);
577 control += KEXEC_CONTROL_PAGE_SIZE;
578 kexec_mark_range(control, crashk_res.end, protect);
581 void arch_kexec_protect_crashkres(void)
583 kexec_mark_crashkres(true);
586 void arch_kexec_unprotect_crashkres(void)
588 kexec_mark_crashkres(false);