1 /*
2  * handle transition of Linux booting another kernel
3  * Copyright (C) 2002-2005 Eric Biederman  <ebiederm@xmission.com>
4  *
5  * This source code is licensed under the GNU General Public License,
6  * Version 2.  See the file COPYING for more details.
7  */
8 
9 #define pr_fmt(fmt)	"kexec: " fmt
10 
11 #include <linux/mm.h>
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>
18 #include <linux/io.h>
19 #include <linux/suspend.h>
20 #include <linux/vmalloc.h>
21 
22 #include <asm/init.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>
30 #include <asm/set_memory.h>
31 
32 #ifdef CONFIG_KEXEC_FILE
33 const struct kexec_file_ops * const kexec_file_loaders[] = {
34 		&kexec_bzImage64_ops,
35 		NULL
36 };
37 #endif
38 
39 static void free_transition_pgtable(struct kimage *image)
40 {
41 	free_page((unsigned long)image->arch.p4d);
42 	image->arch.p4d = NULL;
43 	free_page((unsigned long)image->arch.pud);
44 	image->arch.pud = NULL;
45 	free_page((unsigned long)image->arch.pmd);
46 	image->arch.pmd = NULL;
47 	free_page((unsigned long)image->arch.pte);
48 	image->arch.pte = NULL;
49 }
50 
51 static int init_transition_pgtable(struct kimage *image, pgd_t *pgd)
52 {
53 	p4d_t *p4d;
54 	pud_t *pud;
55 	pmd_t *pmd;
56 	pte_t *pte;
57 	unsigned long vaddr, paddr;
58 	int result = -ENOMEM;
59 
60 	vaddr = (unsigned long)relocate_kernel;
61 	paddr = __pa(page_address(image->control_code_page)+PAGE_SIZE);
62 	pgd += pgd_index(vaddr);
63 	if (!pgd_present(*pgd)) {
64 		p4d = (p4d_t *)get_zeroed_page(GFP_KERNEL);
65 		if (!p4d)
66 			goto err;
67 		image->arch.p4d = p4d;
68 		set_pgd(pgd, __pgd(__pa(p4d) | _KERNPG_TABLE));
69 	}
70 	p4d = p4d_offset(pgd, vaddr);
71 	if (!p4d_present(*p4d)) {
72 		pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
73 		if (!pud)
74 			goto err;
75 		image->arch.pud = pud;
76 		set_p4d(p4d, __p4d(__pa(pud) | _KERNPG_TABLE));
77 	}
78 	pud = pud_offset(p4d, vaddr);
79 	if (!pud_present(*pud)) {
80 		pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
81 		if (!pmd)
82 			goto err;
83 		image->arch.pmd = pmd;
84 		set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
85 	}
86 	pmd = pmd_offset(pud, vaddr);
87 	if (!pmd_present(*pmd)) {
88 		pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
89 		if (!pte)
90 			goto err;
91 		image->arch.pte = pte;
92 		set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
93 	}
94 	pte = pte_offset_kernel(pmd, vaddr);
95 	set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC_NOENC));
96 	return 0;
97 err:
98 	return result;
99 }
100 
101 static void *alloc_pgt_page(void *data)
102 {
103 	struct kimage *image = (struct kimage *)data;
104 	struct page *page;
105 	void *p = NULL;
106 
107 	page = kimage_alloc_control_pages(image, 0);
108 	if (page) {
109 		p = page_address(page);
110 		clear_page(p);
111 	}
112 
113 	return p;
114 }
115 
116 static int init_pgtable(struct kimage *image, unsigned long start_pgtable)
117 {
118 	struct x86_mapping_info info = {
119 		.alloc_pgt_page	= alloc_pgt_page,
120 		.context	= image,
121 		.page_flag	= __PAGE_KERNEL_LARGE_EXEC,
122 		.kernpg_flag	= _KERNPG_TABLE_NOENC,
123 	};
124 	unsigned long mstart, mend;
125 	pgd_t *level4p;
126 	int result;
127 	int i;
128 
129 	level4p = (pgd_t *)__va(start_pgtable);
130 	clear_page(level4p);
131 
132 	if (direct_gbpages)
133 		info.direct_gbpages = true;
134 
135 	for (i = 0; i < nr_pfn_mapped; i++) {
136 		mstart = pfn_mapped[i].start << PAGE_SHIFT;
137 		mend   = pfn_mapped[i].end << PAGE_SHIFT;
138 
139 		result = kernel_ident_mapping_init(&info,
140 						 level4p, mstart, mend);
141 		if (result)
142 			return result;
143 	}
144 
145 	/*
146 	 * segments's mem ranges could be outside 0 ~ max_pfn,
147 	 * for example when jump back to original kernel from kexeced kernel.
148 	 * or first kernel is booted with user mem map, and second kernel
149 	 * could be loaded out of that range.
150 	 */
151 	for (i = 0; i < image->nr_segments; i++) {
152 		mstart = image->segment[i].mem;
153 		mend   = mstart + image->segment[i].memsz;
154 
155 		result = kernel_ident_mapping_init(&info,
156 						 level4p, mstart, mend);
157 
158 		if (result)
159 			return result;
160 	}
161 
162 	return init_transition_pgtable(image, level4p);
163 }
164 
165 static void set_idt(void *newidt, u16 limit)
166 {
167 	struct desc_ptr curidt;
168 
169 	/* x86-64 supports unaliged loads & stores */
170 	curidt.size    = limit;
171 	curidt.address = (unsigned long)newidt;
172 
173 	__asm__ __volatile__ (
174 		"lidtq %0\n"
175 		: : "m" (curidt)
176 		);
177 };
178 
179 
180 static void set_gdt(void *newgdt, u16 limit)
181 {
182 	struct desc_ptr curgdt;
183 
184 	/* x86-64 supports unaligned loads & stores */
185 	curgdt.size    = limit;
186 	curgdt.address = (unsigned long)newgdt;
187 
188 	__asm__ __volatile__ (
189 		"lgdtq %0\n"
190 		: : "m" (curgdt)
191 		);
192 };
193 
194 static void load_segments(void)
195 {
196 	__asm__ __volatile__ (
197 		"\tmovl %0,%%ds\n"
198 		"\tmovl %0,%%es\n"
199 		"\tmovl %0,%%ss\n"
200 		"\tmovl %0,%%fs\n"
201 		"\tmovl %0,%%gs\n"
202 		: : "a" (__KERNEL_DS) : "memory"
203 		);
204 }
205 
206 #ifdef CONFIG_KEXEC_FILE
207 /* Update purgatory as needed after various image segments have been prepared */
208 static int arch_update_purgatory(struct kimage *image)
209 {
210 	int ret = 0;
211 
212 	if (!image->file_mode)
213 		return 0;
214 
215 	/* Setup copying of backup region */
216 	if (image->type == KEXEC_TYPE_CRASH) {
217 		ret = kexec_purgatory_get_set_symbol(image,
218 				"purgatory_backup_dest",
219 				&image->arch.backup_load_addr,
220 				sizeof(image->arch.backup_load_addr), 0);
221 		if (ret)
222 			return ret;
223 
224 		ret = kexec_purgatory_get_set_symbol(image,
225 				"purgatory_backup_src",
226 				&image->arch.backup_src_start,
227 				sizeof(image->arch.backup_src_start), 0);
228 		if (ret)
229 			return ret;
230 
231 		ret = kexec_purgatory_get_set_symbol(image,
232 				"purgatory_backup_sz",
233 				&image->arch.backup_src_sz,
234 				sizeof(image->arch.backup_src_sz), 0);
235 		if (ret)
236 			return ret;
237 	}
238 
239 	return ret;
240 }
241 #else /* !CONFIG_KEXEC_FILE */
242 static inline int arch_update_purgatory(struct kimage *image)
243 {
244 	return 0;
245 }
246 #endif /* CONFIG_KEXEC_FILE */
247 
248 int machine_kexec_prepare(struct kimage *image)
249 {
250 	unsigned long start_pgtable;
251 	int result;
252 
253 	/* Calculate the offsets */
254 	start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
255 
256 	/* Setup the identity mapped 64bit page table */
257 	result = init_pgtable(image, start_pgtable);
258 	if (result)
259 		return result;
260 
261 	/* update purgatory as needed */
262 	result = arch_update_purgatory(image);
263 	if (result)
264 		return result;
265 
266 	return 0;
267 }
268 
269 void machine_kexec_cleanup(struct kimage *image)
270 {
271 	free_transition_pgtable(image);
272 }
273 
274 /*
275  * Do not allocate memory (or fail in any way) in machine_kexec().
276  * We are past the point of no return, committed to rebooting now.
277  */
278 void machine_kexec(struct kimage *image)
279 {
280 	unsigned long page_list[PAGES_NR];
281 	void *control_page;
282 	int save_ftrace_enabled;
283 
284 #ifdef CONFIG_KEXEC_JUMP
285 	if (image->preserve_context)
286 		save_processor_state();
287 #endif
288 
289 	save_ftrace_enabled = __ftrace_enabled_save();
290 
291 	/* Interrupts aren't acceptable while we reboot */
292 	local_irq_disable();
293 	hw_breakpoint_disable();
294 
295 	if (image->preserve_context) {
296 #ifdef CONFIG_X86_IO_APIC
297 		/*
298 		 * We need to put APICs in legacy mode so that we can
299 		 * get timer interrupts in second kernel. kexec/kdump
300 		 * paths already have calls to restore_boot_irq_mode()
301 		 * in one form or other. kexec jump path also need one.
302 		 */
303 		clear_IO_APIC();
304 		restore_boot_irq_mode();
305 #endif
306 	}
307 
308 	control_page = page_address(image->control_code_page) + PAGE_SIZE;
309 	memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
310 
311 	page_list[PA_CONTROL_PAGE] = virt_to_phys(control_page);
312 	page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
313 	page_list[PA_TABLE_PAGE] =
314 	  (unsigned long)__pa(page_address(image->control_code_page));
315 
316 	if (image->type == KEXEC_TYPE_DEFAULT)
317 		page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
318 						<< PAGE_SHIFT);
319 
320 	/*
321 	 * The segment registers are funny things, they have both a
322 	 * visible and an invisible part.  Whenever the visible part is
323 	 * set to a specific selector, the invisible part is loaded
324 	 * with from a table in memory.  At no other time is the
325 	 * descriptor table in memory accessed.
326 	 *
327 	 * I take advantage of this here by force loading the
328 	 * segments, before I zap the gdt with an invalid value.
329 	 */
330 	load_segments();
331 	/*
332 	 * The gdt & idt are now invalid.
333 	 * If you want to load them you must set up your own idt & gdt.
334 	 */
335 	set_gdt(phys_to_virt(0), 0);
336 	set_idt(phys_to_virt(0), 0);
337 
338 	/* now call it */
339 	image->start = relocate_kernel((unsigned long)image->head,
340 				       (unsigned long)page_list,
341 				       image->start,
342 				       image->preserve_context,
343 				       sme_active());
344 
345 #ifdef CONFIG_KEXEC_JUMP
346 	if (image->preserve_context)
347 		restore_processor_state();
348 #endif
349 
350 	__ftrace_enabled_restore(save_ftrace_enabled);
351 }
352 
353 void arch_crash_save_vmcoreinfo(void)
354 {
355 	VMCOREINFO_NUMBER(phys_base);
356 	VMCOREINFO_SYMBOL(init_top_pgt);
357 	vmcoreinfo_append_str("NUMBER(pgtable_l5_enabled)=%d\n",
358 			pgtable_l5_enabled());
359 
360 #ifdef CONFIG_NUMA
361 	VMCOREINFO_SYMBOL(node_data);
362 	VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);
363 #endif
364 	vmcoreinfo_append_str("KERNELOFFSET=%lx\n",
365 			      kaslr_offset());
366 	VMCOREINFO_NUMBER(KERNEL_IMAGE_SIZE);
367 }
368 
369 /* arch-dependent functionality related to kexec file-based syscall */
370 
371 #ifdef CONFIG_KEXEC_FILE
372 void *arch_kexec_kernel_image_load(struct kimage *image)
373 {
374 	vfree(image->arch.elf_headers);
375 	image->arch.elf_headers = NULL;
376 
377 	if (!image->fops || !image->fops->load)
378 		return ERR_PTR(-ENOEXEC);
379 
380 	return image->fops->load(image, image->kernel_buf,
381 				 image->kernel_buf_len, image->initrd_buf,
382 				 image->initrd_buf_len, image->cmdline_buf,
383 				 image->cmdline_buf_len);
384 }
385 
386 /*
387  * Apply purgatory relocations.
388  *
389  * @pi:		Purgatory to be relocated.
390  * @section:	Section relocations applying to.
391  * @relsec:	Section containing RELAs.
392  * @symtabsec:	Corresponding symtab.
393  *
394  * TODO: Some of the code belongs to generic code. Move that in kexec.c.
395  */
396 int arch_kexec_apply_relocations_add(struct purgatory_info *pi,
397 				     Elf_Shdr *section, const Elf_Shdr *relsec,
398 				     const Elf_Shdr *symtabsec)
399 {
400 	unsigned int i;
401 	Elf64_Rela *rel;
402 	Elf64_Sym *sym;
403 	void *location;
404 	unsigned long address, sec_base, value;
405 	const char *strtab, *name, *shstrtab;
406 	const Elf_Shdr *sechdrs;
407 
408 	/* String & section header string table */
409 	sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
410 	strtab = (char *)pi->ehdr + sechdrs[symtabsec->sh_link].sh_offset;
411 	shstrtab = (char *)pi->ehdr + sechdrs[pi->ehdr->e_shstrndx].sh_offset;
412 
413 	rel = (void *)pi->ehdr + relsec->sh_offset;
414 
415 	pr_debug("Applying relocate section %s to %u\n",
416 		 shstrtab + relsec->sh_name, relsec->sh_info);
417 
418 	for (i = 0; i < relsec->sh_size / sizeof(*rel); i++) {
419 
420 		/*
421 		 * rel[i].r_offset contains byte offset from beginning
422 		 * of section to the storage unit affected.
423 		 *
424 		 * This is location to update. This is temporary buffer
425 		 * where section is currently loaded. This will finally be
426 		 * loaded to a different address later, pointed to by
427 		 * ->sh_addr. kexec takes care of moving it
428 		 *  (kexec_load_segment()).
429 		 */
430 		location = pi->purgatory_buf;
431 		location += section->sh_offset;
432 		location += rel[i].r_offset;
433 
434 		/* Final address of the location */
435 		address = section->sh_addr + rel[i].r_offset;
436 
437 		/*
438 		 * rel[i].r_info contains information about symbol table index
439 		 * w.r.t which relocation must be made and type of relocation
440 		 * to apply. ELF64_R_SYM() and ELF64_R_TYPE() macros get
441 		 * these respectively.
442 		 */
443 		sym = (void *)pi->ehdr + symtabsec->sh_offset;
444 		sym += ELF64_R_SYM(rel[i].r_info);
445 
446 		if (sym->st_name)
447 			name = strtab + sym->st_name;
448 		else
449 			name = shstrtab + sechdrs[sym->st_shndx].sh_name;
450 
451 		pr_debug("Symbol: %s info: %02x shndx: %02x value=%llx size: %llx\n",
452 			 name, sym->st_info, sym->st_shndx, sym->st_value,
453 			 sym->st_size);
454 
455 		if (sym->st_shndx == SHN_UNDEF) {
456 			pr_err("Undefined symbol: %s\n", name);
457 			return -ENOEXEC;
458 		}
459 
460 		if (sym->st_shndx == SHN_COMMON) {
461 			pr_err("symbol '%s' in common section\n", name);
462 			return -ENOEXEC;
463 		}
464 
465 		if (sym->st_shndx == SHN_ABS)
466 			sec_base = 0;
467 		else if (sym->st_shndx >= pi->ehdr->e_shnum) {
468 			pr_err("Invalid section %d for symbol %s\n",
469 			       sym->st_shndx, name);
470 			return -ENOEXEC;
471 		} else
472 			sec_base = pi->sechdrs[sym->st_shndx].sh_addr;
473 
474 		value = sym->st_value;
475 		value += sec_base;
476 		value += rel[i].r_addend;
477 
478 		switch (ELF64_R_TYPE(rel[i].r_info)) {
479 		case R_X86_64_NONE:
480 			break;
481 		case R_X86_64_64:
482 			*(u64 *)location = value;
483 			break;
484 		case R_X86_64_32:
485 			*(u32 *)location = value;
486 			if (value != *(u32 *)location)
487 				goto overflow;
488 			break;
489 		case R_X86_64_32S:
490 			*(s32 *)location = value;
491 			if ((s64)value != *(s32 *)location)
492 				goto overflow;
493 			break;
494 		case R_X86_64_PC32:
495 		case R_X86_64_PLT32:
496 			value -= (u64)address;
497 			*(u32 *)location = value;
498 			break;
499 		default:
500 			pr_err("Unknown rela relocation: %llu\n",
501 			       ELF64_R_TYPE(rel[i].r_info));
502 			return -ENOEXEC;
503 		}
504 	}
505 	return 0;
506 
507 overflow:
508 	pr_err("Overflow in relocation type %d value 0x%lx\n",
509 	       (int)ELF64_R_TYPE(rel[i].r_info), value);
510 	return -ENOEXEC;
511 }
512 #endif /* CONFIG_KEXEC_FILE */
513 
514 static int
515 kexec_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, nr_pages);
531 	else
532 		return set_pages_rw(page, nr_pages);
533 }
534 
535 static void kexec_mark_crashkres(bool protect)
536 {
537 	unsigned long control;
538 
539 	kexec_mark_range(crashk_low_res.start, 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(crashk_res.start, control + PAGE_SIZE - 1, protect);
545 	control += KEXEC_CONTROL_PAGE_SIZE;
546 	kexec_mark_range(control, crashk_res.end, protect);
547 }
548 
549 void arch_kexec_protect_crashkres(void)
550 {
551 	kexec_mark_crashkres(true);
552 }
553 
554 void arch_kexec_unprotect_crashkres(void)
555 {
556 	kexec_mark_crashkres(false);
557 }
558 
559 int arch_kexec_post_alloc_pages(void *vaddr, unsigned int pages, gfp_t gfp)
560 {
561 	/*
562 	 * If SME is active we need to be sure that kexec pages are
563 	 * not encrypted because when we boot to the new kernel the
564 	 * pages won't be accessed encrypted (initially).
565 	 */
566 	return set_memory_decrypted((unsigned long)vaddr, pages);
567 }
568 
569 void arch_kexec_pre_free_pages(void *vaddr, unsigned int pages)
570 {
571 	/*
572 	 * If SME is active we need to reset the pages back to being
573 	 * an encrypted mapping before freeing them.
574 	 */
575 	set_memory_encrypted((unsigned long)vaddr, pages);
576 }
577