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 
31 #ifdef CONFIG_ACPI
32 /*
33  * Used while adding mapping for ACPI tables.
34  * Can be reused when other iomem regions need be mapped
35  */
36 struct init_pgtable_data {
37 	struct x86_mapping_info *info;
38 	pgd_t *level4p;
39 };
40 
41 static int mem_region_callback(struct resource *res, void *arg)
42 {
43 	struct init_pgtable_data *data = arg;
44 	unsigned long mstart, mend;
45 
46 	mstart = res->start;
47 	mend = mstart + resource_size(res) - 1;
48 
49 	return kernel_ident_mapping_init(data->info, data->level4p, mstart, mend);
50 }
51 
52 static int
53 map_acpi_tables(struct x86_mapping_info *info, pgd_t *level4p)
54 {
55 	struct init_pgtable_data data;
56 	unsigned long flags;
57 	int ret;
58 
59 	data.info = info;
60 	data.level4p = level4p;
61 	flags = IORESOURCE_MEM | IORESOURCE_BUSY;
62 
63 	ret = walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1,
64 				  &data, mem_region_callback);
65 	if (ret && ret != -EINVAL)
66 		return ret;
67 
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)
72 		return ret;
73 
74 	return 0;
75 }
76 #else
77 static int map_acpi_tables(struct x86_mapping_info *info, pgd_t *level4p) { return 0; }
78 #endif
79 
80 #ifdef CONFIG_KEXEC_FILE
81 const struct kexec_file_ops * const kexec_file_loaders[] = {
82 		&kexec_bzImage64_ops,
83 		NULL
84 };
85 #endif
86 
87 static int
88 map_efi_systab(struct x86_mapping_info *info, pgd_t *level4p)
89 {
90 #ifdef CONFIG_EFI
91 	unsigned long mstart, mend;
92 
93 	if (!efi_enabled(EFI_BOOT))
94 		return 0;
95 
96 	mstart = (boot_params.efi_info.efi_systab |
97 			((u64)boot_params.efi_info.efi_systab_hi<<32));
98 
99 	if (efi_enabled(EFI_64BIT))
100 		mend = mstart + sizeof(efi_system_table_64_t);
101 	else
102 		mend = mstart + sizeof(efi_system_table_32_t);
103 
104 	if (!mstart)
105 		return 0;
106 
107 	return kernel_ident_mapping_init(info, level4p, mstart, mend);
108 #endif
109 	return 0;
110 }
111 
112 static void free_transition_pgtable(struct kimage *image)
113 {
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;
122 }
123 
124 static int init_transition_pgtable(struct kimage *image, pgd_t *pgd)
125 {
126 	pgprot_t prot = PAGE_KERNEL_EXEC_NOENC;
127 	unsigned long vaddr, paddr;
128 	int result = -ENOMEM;
129 	p4d_t *p4d;
130 	pud_t *pud;
131 	pmd_t *pmd;
132 	pte_t *pte;
133 
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);
139 		if (!p4d)
140 			goto err;
141 		image->arch.p4d = p4d;
142 		set_pgd(pgd, __pgd(__pa(p4d) | _KERNPG_TABLE));
143 	}
144 	p4d = p4d_offset(pgd, vaddr);
145 	if (!p4d_present(*p4d)) {
146 		pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
147 		if (!pud)
148 			goto err;
149 		image->arch.pud = pud;
150 		set_p4d(p4d, __p4d(__pa(pud) | _KERNPG_TABLE));
151 	}
152 	pud = pud_offset(p4d, vaddr);
153 	if (!pud_present(*pud)) {
154 		pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
155 		if (!pmd)
156 			goto err;
157 		image->arch.pmd = pmd;
158 		set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
159 	}
160 	pmd = pmd_offset(pud, vaddr);
161 	if (!pmd_present(*pmd)) {
162 		pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
163 		if (!pte)
164 			goto err;
165 		image->arch.pte = pte;
166 		set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
167 	}
168 	pte = pte_offset_kernel(pmd, vaddr);
169 
170 	if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT))
171 		prot = PAGE_KERNEL_EXEC;
172 
173 	set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, prot));
174 	return 0;
175 err:
176 	return result;
177 }
178 
179 static void *alloc_pgt_page(void *data)
180 {
181 	struct kimage *image = (struct kimage *)data;
182 	struct page *page;
183 	void *p = NULL;
184 
185 	page = kimage_alloc_control_pages(image, 0);
186 	if (page) {
187 		p = page_address(page);
188 		clear_page(p);
189 	}
190 
191 	return p;
192 }
193 
194 static int init_pgtable(struct kimage *image, unsigned long start_pgtable)
195 {
196 	struct x86_mapping_info info = {
197 		.alloc_pgt_page	= alloc_pgt_page,
198 		.context	= image,
199 		.page_flag	= __PAGE_KERNEL_LARGE_EXEC,
200 		.kernpg_flag	= _KERNPG_TABLE_NOENC,
201 	};
202 	unsigned long mstart, mend;
203 	pgd_t *level4p;
204 	int result;
205 	int i;
206 
207 	level4p = (pgd_t *)__va(start_pgtable);
208 	clear_page(level4p);
209 
210 	if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT)) {
211 		info.page_flag   |= _PAGE_ENC;
212 		info.kernpg_flag |= _PAGE_ENC;
213 	}
214 
215 	if (direct_gbpages)
216 		info.direct_gbpages = true;
217 
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;
221 
222 		result = kernel_ident_mapping_init(&info,
223 						 level4p, mstart, mend);
224 		if (result)
225 			return result;
226 	}
227 
228 	/*
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.
233 	 */
234 	for (i = 0; i < image->nr_segments; i++) {
235 		mstart = image->segment[i].mem;
236 		mend   = mstart + image->segment[i].memsz;
237 
238 		result = kernel_ident_mapping_init(&info,
239 						 level4p, mstart, mend);
240 
241 		if (result)
242 			return result;
243 	}
244 
245 	/*
246 	 * Prepare EFI systab and ACPI tables for kexec kernel since they are
247 	 * not covered by pfn_mapped.
248 	 */
249 	result = map_efi_systab(&info, level4p);
250 	if (result)
251 		return result;
252 
253 	result = map_acpi_tables(&info, level4p);
254 	if (result)
255 		return result;
256 
257 	return init_transition_pgtable(image, level4p);
258 }
259 
260 static void load_segments(void)
261 {
262 	__asm__ __volatile__ (
263 		"\tmovl %0,%%ds\n"
264 		"\tmovl %0,%%es\n"
265 		"\tmovl %0,%%ss\n"
266 		"\tmovl %0,%%fs\n"
267 		"\tmovl %0,%%gs\n"
268 		: : "a" (__KERNEL_DS) : "memory"
269 		);
270 }
271 
272 int machine_kexec_prepare(struct kimage *image)
273 {
274 	unsigned long start_pgtable;
275 	int result;
276 
277 	/* Calculate the offsets */
278 	start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
279 
280 	/* Setup the identity mapped 64bit page table */
281 	result = init_pgtable(image, start_pgtable);
282 	if (result)
283 		return result;
284 
285 	return 0;
286 }
287 
288 void machine_kexec_cleanup(struct kimage *image)
289 {
290 	free_transition_pgtable(image);
291 }
292 
293 /*
294  * Do not allocate memory (or fail in any way) in machine_kexec().
295  * We are past the point of no return, committed to rebooting now.
296  */
297 void machine_kexec(struct kimage *image)
298 {
299 	unsigned long page_list[PAGES_NR];
300 	void *control_page;
301 	int save_ftrace_enabled;
302 
303 #ifdef CONFIG_KEXEC_JUMP
304 	if (image->preserve_context)
305 		save_processor_state();
306 #endif
307 
308 	save_ftrace_enabled = __ftrace_enabled_save();
309 
310 	/* Interrupts aren't acceptable while we reboot */
311 	local_irq_disable();
312 	hw_breakpoint_disable();
313 
314 	if (image->preserve_context) {
315 #ifdef CONFIG_X86_IO_APIC
316 		/*
317 		 * We need to put APICs in legacy mode so that we can
318 		 * get timer interrupts in second kernel. kexec/kdump
319 		 * paths already have calls to restore_boot_irq_mode()
320 		 * in one form or other. kexec jump path also need one.
321 		 */
322 		clear_IO_APIC();
323 		restore_boot_irq_mode();
324 #endif
325 	}
326 
327 	control_page = page_address(image->control_code_page) + PAGE_SIZE;
328 	memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
329 
330 	page_list[PA_CONTROL_PAGE] = virt_to_phys(control_page);
331 	page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
332 	page_list[PA_TABLE_PAGE] =
333 	  (unsigned long)__pa(page_address(image->control_code_page));
334 
335 	if (image->type == KEXEC_TYPE_DEFAULT)
336 		page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
337 						<< PAGE_SHIFT);
338 
339 	/*
340 	 * The segment registers are funny things, they have both a
341 	 * visible and an invisible part.  Whenever the visible part is
342 	 * set to a specific selector, the invisible part is loaded
343 	 * with from a table in memory.  At no other time is the
344 	 * descriptor table in memory accessed.
345 	 *
346 	 * I take advantage of this here by force loading the
347 	 * segments, before I zap the gdt with an invalid value.
348 	 */
349 	load_segments();
350 	/*
351 	 * The gdt & idt are now invalid.
352 	 * If you want to load them you must set up your own idt & gdt.
353 	 */
354 	native_idt_invalidate();
355 	native_gdt_invalidate();
356 
357 	/* now call it */
358 	image->start = relocate_kernel((unsigned long)image->head,
359 				       (unsigned long)page_list,
360 				       image->start,
361 				       image->preserve_context,
362 				       cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT));
363 
364 #ifdef CONFIG_KEXEC_JUMP
365 	if (image->preserve_context)
366 		restore_processor_state();
367 #endif
368 
369 	__ftrace_enabled_restore(save_ftrace_enabled);
370 }
371 
372 /* arch-dependent functionality related to kexec file-based syscall */
373 
374 #ifdef CONFIG_KEXEC_FILE
375 void *arch_kexec_kernel_image_load(struct kimage *image)
376 {
377 	vfree(image->elf_headers);
378 	image->elf_headers = NULL;
379 
380 	if (!image->fops || !image->fops->load)
381 		return ERR_PTR(-ENOEXEC);
382 
383 	return image->fops->load(image, image->kernel_buf,
384 				 image->kernel_buf_len, image->initrd_buf,
385 				 image->initrd_buf_len, image->cmdline_buf,
386 				 image->cmdline_buf_len);
387 }
388 
389 /*
390  * Apply purgatory relocations.
391  *
392  * @pi:		Purgatory to be relocated.
393  * @section:	Section relocations applying to.
394  * @relsec:	Section containing RELAs.
395  * @symtabsec:	Corresponding symtab.
396  *
397  * TODO: Some of the code belongs to generic code. Move that in kexec.c.
398  */
399 int arch_kexec_apply_relocations_add(struct purgatory_info *pi,
400 				     Elf_Shdr *section, const Elf_Shdr *relsec,
401 				     const Elf_Shdr *symtabsec)
402 {
403 	unsigned int i;
404 	Elf64_Rela *rel;
405 	Elf64_Sym *sym;
406 	void *location;
407 	unsigned long address, sec_base, value;
408 	const char *strtab, *name, *shstrtab;
409 	const Elf_Shdr *sechdrs;
410 
411 	/* String & section header string table */
412 	sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
413 	strtab = (char *)pi->ehdr + sechdrs[symtabsec->sh_link].sh_offset;
414 	shstrtab = (char *)pi->ehdr + sechdrs[pi->ehdr->e_shstrndx].sh_offset;
415 
416 	rel = (void *)pi->ehdr + relsec->sh_offset;
417 
418 	pr_debug("Applying relocate section %s to %u\n",
419 		 shstrtab + relsec->sh_name, relsec->sh_info);
420 
421 	for (i = 0; i < relsec->sh_size / sizeof(*rel); i++) {
422 
423 		/*
424 		 * rel[i].r_offset contains byte offset from beginning
425 		 * of section to the storage unit affected.
426 		 *
427 		 * This is location to update. This is temporary buffer
428 		 * where section is currently loaded. This will finally be
429 		 * loaded to a different address later, pointed to by
430 		 * ->sh_addr. kexec takes care of moving it
431 		 *  (kexec_load_segment()).
432 		 */
433 		location = pi->purgatory_buf;
434 		location += section->sh_offset;
435 		location += rel[i].r_offset;
436 
437 		/* Final address of the location */
438 		address = section->sh_addr + rel[i].r_offset;
439 
440 		/*
441 		 * rel[i].r_info contains information about symbol table index
442 		 * w.r.t which relocation must be made and type of relocation
443 		 * to apply. ELF64_R_SYM() and ELF64_R_TYPE() macros get
444 		 * these respectively.
445 		 */
446 		sym = (void *)pi->ehdr + symtabsec->sh_offset;
447 		sym += ELF64_R_SYM(rel[i].r_info);
448 
449 		if (sym->st_name)
450 			name = strtab + sym->st_name;
451 		else
452 			name = shstrtab + sechdrs[sym->st_shndx].sh_name;
453 
454 		pr_debug("Symbol: %s info: %02x shndx: %02x value=%llx size: %llx\n",
455 			 name, sym->st_info, sym->st_shndx, sym->st_value,
456 			 sym->st_size);
457 
458 		if (sym->st_shndx == SHN_UNDEF) {
459 			pr_err("Undefined symbol: %s\n", name);
460 			return -ENOEXEC;
461 		}
462 
463 		if (sym->st_shndx == SHN_COMMON) {
464 			pr_err("symbol '%s' in common section\n", name);
465 			return -ENOEXEC;
466 		}
467 
468 		if (sym->st_shndx == SHN_ABS)
469 			sec_base = 0;
470 		else if (sym->st_shndx >= pi->ehdr->e_shnum) {
471 			pr_err("Invalid section %d for symbol %s\n",
472 			       sym->st_shndx, name);
473 			return -ENOEXEC;
474 		} else
475 			sec_base = pi->sechdrs[sym->st_shndx].sh_addr;
476 
477 		value = sym->st_value;
478 		value += sec_base;
479 		value += rel[i].r_addend;
480 
481 		switch (ELF64_R_TYPE(rel[i].r_info)) {
482 		case R_X86_64_NONE:
483 			break;
484 		case R_X86_64_64:
485 			*(u64 *)location = value;
486 			break;
487 		case R_X86_64_32:
488 			*(u32 *)location = value;
489 			if (value != *(u32 *)location)
490 				goto overflow;
491 			break;
492 		case R_X86_64_32S:
493 			*(s32 *)location = value;
494 			if ((s64)value != *(s32 *)location)
495 				goto overflow;
496 			break;
497 		case R_X86_64_PC32:
498 		case R_X86_64_PLT32:
499 			value -= (u64)address;
500 			*(u32 *)location = value;
501 			break;
502 		default:
503 			pr_err("Unknown rela relocation: %llu\n",
504 			       ELF64_R_TYPE(rel[i].r_info));
505 			return -ENOEXEC;
506 		}
507 	}
508 	return 0;
509 
510 overflow:
511 	pr_err("Overflow in relocation type %d value 0x%lx\n",
512 	       (int)ELF64_R_TYPE(rel[i].r_info), value);
513 	return -ENOEXEC;
514 }
515 #endif /* CONFIG_KEXEC_FILE */
516 
517 static int
518 kexec_mark_range(unsigned long start, unsigned long end, bool protect)
519 {
520 	struct page *page;
521 	unsigned int nr_pages;
522 
523 	/*
524 	 * For physical range: [start, end]. We must skip the unassigned
525 	 * crashk resource with zero-valued "end" member.
526 	 */
527 	if (!end || start > end)
528 		return 0;
529 
530 	page = pfn_to_page(start >> PAGE_SHIFT);
531 	nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
532 	if (protect)
533 		return set_pages_ro(page, nr_pages);
534 	else
535 		return set_pages_rw(page, nr_pages);
536 }
537 
538 static void kexec_mark_crashkres(bool protect)
539 {
540 	unsigned long control;
541 
542 	kexec_mark_range(crashk_low_res.start, crashk_low_res.end, protect);
543 
544 	/* Don't touch the control code page used in crash_kexec().*/
545 	control = PFN_PHYS(page_to_pfn(kexec_crash_image->control_code_page));
546 	/* Control code page is located in the 2nd page. */
547 	kexec_mark_range(crashk_res.start, control + PAGE_SIZE - 1, protect);
548 	control += KEXEC_CONTROL_PAGE_SIZE;
549 	kexec_mark_range(control, crashk_res.end, protect);
550 }
551 
552 void arch_kexec_protect_crashkres(void)
553 {
554 	kexec_mark_crashkres(true);
555 }
556 
557 void arch_kexec_unprotect_crashkres(void)
558 {
559 	kexec_mark_crashkres(false);
560 }
561 
562 /*
563  * During a traditional boot under SME, SME will encrypt the kernel,
564  * so the SME kexec kernel also needs to be un-encrypted in order to
565  * replicate a normal SME boot.
566  *
567  * During a traditional boot under SEV, the kernel has already been
568  * loaded encrypted, so the SEV kexec kernel needs to be encrypted in
569  * order to replicate a normal SEV boot.
570  */
571 int arch_kexec_post_alloc_pages(void *vaddr, unsigned int pages, gfp_t gfp)
572 {
573 	if (!cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT))
574 		return 0;
575 
576 	/*
577 	 * If host memory encryption is active we need to be sure that kexec
578 	 * pages are not encrypted because when we boot to the new kernel the
579 	 * pages won't be accessed encrypted (initially).
580 	 */
581 	return set_memory_decrypted((unsigned long)vaddr, pages);
582 }
583 
584 void arch_kexec_pre_free_pages(void *vaddr, unsigned int pages)
585 {
586 	if (!cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT))
587 		return;
588 
589 	/*
590 	 * If host memory encryption is active we need to reset the pages back
591 	 * to being an encrypted mapping before freeing them.
592 	 */
593 	set_memory_encrypted((unsigned long)vaddr, pages);
594 }
595