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