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 
31 #ifdef CONFIG_KEXEC_FILE
32 static struct kexec_file_ops *kexec_file_loaders[] = {
33 		&kexec_bzImage64_ops,
34 };
35 #endif
36 
37 static void free_transition_pgtable(struct kimage *image)
38 {
39 	free_page((unsigned long)image->arch.pud);
40 	free_page((unsigned long)image->arch.pmd);
41 	free_page((unsigned long)image->arch.pte);
42 }
43 
44 static int init_transition_pgtable(struct kimage *image, pgd_t *pgd)
45 {
46 	pud_t *pud;
47 	pmd_t *pmd;
48 	pte_t *pte;
49 	unsigned long vaddr, paddr;
50 	int result = -ENOMEM;
51 
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);
57 		if (!pud)
58 			goto err;
59 		image->arch.pud = pud;
60 		set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE));
61 	}
62 	pud = pud_offset(pgd, vaddr);
63 	if (!pud_present(*pud)) {
64 		pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
65 		if (!pmd)
66 			goto err;
67 		image->arch.pmd = pmd;
68 		set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
69 	}
70 	pmd = pmd_offset(pud, vaddr);
71 	if (!pmd_present(*pmd)) {
72 		pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
73 		if (!pte)
74 			goto err;
75 		image->arch.pte = pte;
76 		set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
77 	}
78 	pte = pte_offset_kernel(pmd, vaddr);
79 	set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC));
80 	return 0;
81 err:
82 	free_transition_pgtable(image);
83 	return result;
84 }
85 
86 static void *alloc_pgt_page(void *data)
87 {
88 	struct kimage *image = (struct kimage *)data;
89 	struct page *page;
90 	void *p = NULL;
91 
92 	page = kimage_alloc_control_pages(image, 0);
93 	if (page) {
94 		p = page_address(page);
95 		clear_page(p);
96 	}
97 
98 	return p;
99 }
100 
101 static int init_pgtable(struct kimage *image, unsigned long start_pgtable)
102 {
103 	struct x86_mapping_info info = {
104 		.alloc_pgt_page	= alloc_pgt_page,
105 		.context	= image,
106 		.pmd_flag	= __PAGE_KERNEL_LARGE_EXEC,
107 	};
108 	unsigned long mstart, mend;
109 	pgd_t *level4p;
110 	int result;
111 	int i;
112 
113 	level4p = (pgd_t *)__va(start_pgtable);
114 	clear_page(level4p);
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;
118 
119 		result = kernel_ident_mapping_init(&info,
120 						 level4p, mstart, mend);
121 		if (result)
122 			return result;
123 	}
124 
125 	/*
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.
130 	 */
131 	for (i = 0; i < image->nr_segments; i++) {
132 		mstart = image->segment[i].mem;
133 		mend   = mstart + image->segment[i].memsz;
134 
135 		result = kernel_ident_mapping_init(&info,
136 						 level4p, mstart, mend);
137 
138 		if (result)
139 			return result;
140 	}
141 
142 	return init_transition_pgtable(image, level4p);
143 }
144 
145 static void set_idt(void *newidt, u16 limit)
146 {
147 	struct desc_ptr curidt;
148 
149 	/* x86-64 supports unaliged loads & stores */
150 	curidt.size    = limit;
151 	curidt.address = (unsigned long)newidt;
152 
153 	__asm__ __volatile__ (
154 		"lidtq %0\n"
155 		: : "m" (curidt)
156 		);
157 };
158 
159 
160 static void set_gdt(void *newgdt, u16 limit)
161 {
162 	struct desc_ptr curgdt;
163 
164 	/* x86-64 supports unaligned loads & stores */
165 	curgdt.size    = limit;
166 	curgdt.address = (unsigned long)newgdt;
167 
168 	__asm__ __volatile__ (
169 		"lgdtq %0\n"
170 		: : "m" (curgdt)
171 		);
172 };
173 
174 static void load_segments(void)
175 {
176 	__asm__ __volatile__ (
177 		"\tmovl %0,%%ds\n"
178 		"\tmovl %0,%%es\n"
179 		"\tmovl %0,%%ss\n"
180 		"\tmovl %0,%%fs\n"
181 		"\tmovl %0,%%gs\n"
182 		: : "a" (__KERNEL_DS) : "memory"
183 		);
184 }
185 
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)
189 {
190 	int ret = 0;
191 
192 	if (!image->file_mode)
193 		return 0;
194 
195 	/* Setup copying of backup region */
196 	if (image->type == KEXEC_TYPE_CRASH) {
197 		ret = kexec_purgatory_get_set_symbol(image, "backup_dest",
198 				&image->arch.backup_load_addr,
199 				sizeof(image->arch.backup_load_addr), 0);
200 		if (ret)
201 			return ret;
202 
203 		ret = kexec_purgatory_get_set_symbol(image, "backup_src",
204 				&image->arch.backup_src_start,
205 				sizeof(image->arch.backup_src_start), 0);
206 		if (ret)
207 			return ret;
208 
209 		ret = kexec_purgatory_get_set_symbol(image, "backup_sz",
210 				&image->arch.backup_src_sz,
211 				sizeof(image->arch.backup_src_sz), 0);
212 		if (ret)
213 			return ret;
214 	}
215 
216 	return ret;
217 }
218 #else /* !CONFIG_KEXEC_FILE */
219 static inline int arch_update_purgatory(struct kimage *image)
220 {
221 	return 0;
222 }
223 #endif /* CONFIG_KEXEC_FILE */
224 
225 int machine_kexec_prepare(struct kimage *image)
226 {
227 	unsigned long start_pgtable;
228 	int result;
229 
230 	/* Calculate the offsets */
231 	start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
232 
233 	/* Setup the identity mapped 64bit page table */
234 	result = init_pgtable(image, start_pgtable);
235 	if (result)
236 		return result;
237 
238 	/* update purgatory as needed */
239 	result = arch_update_purgatory(image);
240 	if (result)
241 		return result;
242 
243 	return 0;
244 }
245 
246 void machine_kexec_cleanup(struct kimage *image)
247 {
248 	free_transition_pgtable(image);
249 }
250 
251 /*
252  * Do not allocate memory (or fail in any way) in machine_kexec().
253  * We are past the point of no return, committed to rebooting now.
254  */
255 void machine_kexec(struct kimage *image)
256 {
257 	unsigned long page_list[PAGES_NR];
258 	void *control_page;
259 	int save_ftrace_enabled;
260 
261 #ifdef CONFIG_KEXEC_JUMP
262 	if (image->preserve_context)
263 		save_processor_state();
264 #endif
265 
266 	save_ftrace_enabled = __ftrace_enabled_save();
267 
268 	/* Interrupts aren't acceptable while we reboot */
269 	local_irq_disable();
270 	hw_breakpoint_disable();
271 
272 	if (image->preserve_context) {
273 #ifdef CONFIG_X86_IO_APIC
274 		/*
275 		 * We need to put APICs in legacy mode so that we can
276 		 * get timer interrupts in second kernel. kexec/kdump
277 		 * paths already have calls to disable_IO_APIC() in
278 		 * one form or other. kexec jump path also need
279 		 * one.
280 		 */
281 		disable_IO_APIC();
282 #endif
283 	}
284 
285 	control_page = page_address(image->control_code_page) + PAGE_SIZE;
286 	memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
287 
288 	page_list[PA_CONTROL_PAGE] = virt_to_phys(control_page);
289 	page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
290 	page_list[PA_TABLE_PAGE] =
291 	  (unsigned long)__pa(page_address(image->control_code_page));
292 
293 	if (image->type == KEXEC_TYPE_DEFAULT)
294 		page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
295 						<< PAGE_SHIFT);
296 
297 	/*
298 	 * The segment registers are funny things, they have both a
299 	 * visible and an invisible part.  Whenever the visible part is
300 	 * set to a specific selector, the invisible part is loaded
301 	 * with from a table in memory.  At no other time is the
302 	 * descriptor table in memory accessed.
303 	 *
304 	 * I take advantage of this here by force loading the
305 	 * segments, before I zap the gdt with an invalid value.
306 	 */
307 	load_segments();
308 	/*
309 	 * The gdt & idt are now invalid.
310 	 * If you want to load them you must set up your own idt & gdt.
311 	 */
312 	set_gdt(phys_to_virt(0), 0);
313 	set_idt(phys_to_virt(0), 0);
314 
315 	/* now call it */
316 	image->start = relocate_kernel((unsigned long)image->head,
317 				       (unsigned long)page_list,
318 				       image->start,
319 				       image->preserve_context);
320 
321 #ifdef CONFIG_KEXEC_JUMP
322 	if (image->preserve_context)
323 		restore_processor_state();
324 #endif
325 
326 	__ftrace_enabled_restore(save_ftrace_enabled);
327 }
328 
329 void arch_crash_save_vmcoreinfo(void)
330 {
331 	VMCOREINFO_SYMBOL(phys_base);
332 	VMCOREINFO_SYMBOL(init_level4_pgt);
333 
334 #ifdef CONFIG_NUMA
335 	VMCOREINFO_SYMBOL(node_data);
336 	VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);
337 #endif
338 	vmcoreinfo_append_str("KERNELOFFSET=%lx\n",
339 			      kaslr_offset());
340 }
341 
342 /* arch-dependent functionality related to kexec file-based syscall */
343 
344 #ifdef CONFIG_KEXEC_FILE
345 int arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
346 				  unsigned long buf_len)
347 {
348 	int i, ret = -ENOEXEC;
349 	struct kexec_file_ops *fops;
350 
351 	for (i = 0; i < ARRAY_SIZE(kexec_file_loaders); i++) {
352 		fops = kexec_file_loaders[i];
353 		if (!fops || !fops->probe)
354 			continue;
355 
356 		ret = fops->probe(buf, buf_len);
357 		if (!ret) {
358 			image->fops = fops;
359 			return ret;
360 		}
361 	}
362 
363 	return ret;
364 }
365 
366 void *arch_kexec_kernel_image_load(struct kimage *image)
367 {
368 	vfree(image->arch.elf_headers);
369 	image->arch.elf_headers = NULL;
370 
371 	if (!image->fops || !image->fops->load)
372 		return ERR_PTR(-ENOEXEC);
373 
374 	return image->fops->load(image, image->kernel_buf,
375 				 image->kernel_buf_len, image->initrd_buf,
376 				 image->initrd_buf_len, image->cmdline_buf,
377 				 image->cmdline_buf_len);
378 }
379 
380 int arch_kimage_file_post_load_cleanup(struct kimage *image)
381 {
382 	if (!image->fops || !image->fops->cleanup)
383 		return 0;
384 
385 	return image->fops->cleanup(image->image_loader_data);
386 }
387 
388 #ifdef CONFIG_KEXEC_VERIFY_SIG
389 int arch_kexec_kernel_verify_sig(struct kimage *image, void *kernel,
390 				 unsigned long kernel_len)
391 {
392 	if (!image->fops || !image->fops->verify_sig) {
393 		pr_debug("kernel loader does not support signature verification.");
394 		return -EKEYREJECTED;
395 	}
396 
397 	return image->fops->verify_sig(kernel, kernel_len);
398 }
399 #endif
400 
401 /*
402  * Apply purgatory relocations.
403  *
404  * ehdr: Pointer to elf headers
405  * sechdrs: Pointer to section headers.
406  * relsec: section index of SHT_RELA section.
407  *
408  * TODO: Some of the code belongs to generic code. Move that in kexec.c.
409  */
410 int arch_kexec_apply_relocations_add(const Elf64_Ehdr *ehdr,
411 				     Elf64_Shdr *sechdrs, unsigned int relsec)
412 {
413 	unsigned int i;
414 	Elf64_Rela *rel;
415 	Elf64_Sym *sym;
416 	void *location;
417 	Elf64_Shdr *section, *symtabsec;
418 	unsigned long address, sec_base, value;
419 	const char *strtab, *name, *shstrtab;
420 
421 	/*
422 	 * ->sh_offset has been modified to keep the pointer to section
423 	 * contents in memory
424 	 */
425 	rel = (void *)sechdrs[relsec].sh_offset;
426 
427 	/* Section to which relocations apply */
428 	section = &sechdrs[sechdrs[relsec].sh_info];
429 
430 	pr_debug("Applying relocate section %u to %u\n", relsec,
431 		 sechdrs[relsec].sh_info);
432 
433 	/* Associated symbol table */
434 	symtabsec = &sechdrs[sechdrs[relsec].sh_link];
435 
436 	/* String table */
437 	if (symtabsec->sh_link >= ehdr->e_shnum) {
438 		/* Invalid strtab section number */
439 		pr_err("Invalid string table section index %d\n",
440 		       symtabsec->sh_link);
441 		return -ENOEXEC;
442 	}
443 
444 	strtab = (char *)sechdrs[symtabsec->sh_link].sh_offset;
445 
446 	/* section header string table */
447 	shstrtab = (char *)sechdrs[ehdr->e_shstrndx].sh_offset;
448 
449 	for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rel); i++) {
450 
451 		/*
452 		 * rel[i].r_offset contains byte offset from beginning
453 		 * of section to the storage unit affected.
454 		 *
455 		 * This is location to update (->sh_offset). This is temporary
456 		 * buffer where section is currently loaded. This will finally
457 		 * be loaded to a different address later, pointed to by
458 		 * ->sh_addr. kexec takes care of moving it
459 		 *  (kexec_load_segment()).
460 		 */
461 		location = (void *)(section->sh_offset + rel[i].r_offset);
462 
463 		/* Final address of the location */
464 		address = section->sh_addr + rel[i].r_offset;
465 
466 		/*
467 		 * rel[i].r_info contains information about symbol table index
468 		 * w.r.t which relocation must be made and type of relocation
469 		 * to apply. ELF64_R_SYM() and ELF64_R_TYPE() macros get
470 		 * these respectively.
471 		 */
472 		sym = (Elf64_Sym *)symtabsec->sh_offset +
473 				ELF64_R_SYM(rel[i].r_info);
474 
475 		if (sym->st_name)
476 			name = strtab + sym->st_name;
477 		else
478 			name = shstrtab + sechdrs[sym->st_shndx].sh_name;
479 
480 		pr_debug("Symbol: %s info: %02x shndx: %02x value=%llx size: %llx\n",
481 			 name, sym->st_info, sym->st_shndx, sym->st_value,
482 			 sym->st_size);
483 
484 		if (sym->st_shndx == SHN_UNDEF) {
485 			pr_err("Undefined symbol: %s\n", name);
486 			return -ENOEXEC;
487 		}
488 
489 		if (sym->st_shndx == SHN_COMMON) {
490 			pr_err("symbol '%s' in common section\n", name);
491 			return -ENOEXEC;
492 		}
493 
494 		if (sym->st_shndx == SHN_ABS)
495 			sec_base = 0;
496 		else if (sym->st_shndx >= ehdr->e_shnum) {
497 			pr_err("Invalid section %d for symbol %s\n",
498 			       sym->st_shndx, name);
499 			return -ENOEXEC;
500 		} else
501 			sec_base = sechdrs[sym->st_shndx].sh_addr;
502 
503 		value = sym->st_value;
504 		value += sec_base;
505 		value += rel[i].r_addend;
506 
507 		switch (ELF64_R_TYPE(rel[i].r_info)) {
508 		case R_X86_64_NONE:
509 			break;
510 		case R_X86_64_64:
511 			*(u64 *)location = value;
512 			break;
513 		case R_X86_64_32:
514 			*(u32 *)location = value;
515 			if (value != *(u32 *)location)
516 				goto overflow;
517 			break;
518 		case R_X86_64_32S:
519 			*(s32 *)location = value;
520 			if ((s64)value != *(s32 *)location)
521 				goto overflow;
522 			break;
523 		case R_X86_64_PC32:
524 			value -= (u64)address;
525 			*(u32 *)location = value;
526 			break;
527 		default:
528 			pr_err("Unknown rela relocation: %llu\n",
529 			       ELF64_R_TYPE(rel[i].r_info));
530 			return -ENOEXEC;
531 		}
532 	}
533 	return 0;
534 
535 overflow:
536 	pr_err("Overflow in relocation type %d value 0x%lx\n",
537 	       (int)ELF64_R_TYPE(rel[i].r_info), value);
538 	return -ENOEXEC;
539 }
540 #endif /* CONFIG_KEXEC_FILE */
541 
542 static int
543 kexec_mark_range(unsigned long start, unsigned long end, bool protect)
544 {
545 	struct page *page;
546 	unsigned int nr_pages;
547 
548 	/*
549 	 * For physical range: [start, end]. We must skip the unassigned
550 	 * crashk resource with zero-valued "end" member.
551 	 */
552 	if (!end || start > end)
553 		return 0;
554 
555 	page = pfn_to_page(start >> PAGE_SHIFT);
556 	nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
557 	if (protect)
558 		return set_pages_ro(page, nr_pages);
559 	else
560 		return set_pages_rw(page, nr_pages);
561 }
562 
563 static void kexec_mark_crashkres(bool protect)
564 {
565 	unsigned long control;
566 
567 	kexec_mark_range(crashk_low_res.start, crashk_low_res.end, protect);
568 
569 	/* Don't touch the control code page used in crash_kexec().*/
570 	control = PFN_PHYS(page_to_pfn(kexec_crash_image->control_code_page));
571 	/* Control code page is located in the 2nd page. */
572 	kexec_mark_range(crashk_res.start, control + PAGE_SIZE - 1, protect);
573 	control += KEXEC_CONTROL_PAGE_SIZE;
574 	kexec_mark_range(control, crashk_res.end, protect);
575 }
576 
577 void arch_kexec_protect_crashkres(void)
578 {
579 	kexec_mark_crashkres(true);
580 }
581 
582 void arch_kexec_unprotect_crashkres(void)
583 {
584 	kexec_mark_crashkres(false);
585 }
586