Linux Audio

Check our new training course

Loading...
v4.17
  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
 33const struct kexec_file_ops * const kexec_file_loaders[] = {
 34		&kexec_bzImage64_ops,
 35		NULL
 36};
 37#endif
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 38
 39static 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
 51static 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;
 97err:
 
 98	return result;
 99}
100
101static 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
116static 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
165static 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
180static 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
194static 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 */
208static 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 */
242static inline int arch_update_purgatory(struct kimage *image)
243{
244	return 0;
245}
246#endif /* CONFIG_KEXEC_FILE */
247
248int 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
269void 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 */
278void 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
353void arch_crash_save_vmcoreinfo(void)
354{
355	VMCOREINFO_NUMBER(phys_base);
356	VMCOREINFO_SYMBOL(init_top_pgt);
357	VMCOREINFO_NUMBER(pgtable_l5_enabled);
358
359#ifdef CONFIG_NUMA
360	VMCOREINFO_SYMBOL(node_data);
361	VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);
362#endif
363	vmcoreinfo_append_str("KERNELOFFSET=%lx\n",
364			      kaslr_offset());
365	VMCOREINFO_NUMBER(KERNEL_IMAGE_SIZE);
366}
367
368/* arch-dependent functionality related to kexec file-based syscall */
369
370#ifdef CONFIG_KEXEC_FILE
371void *arch_kexec_kernel_image_load(struct kimage *image)
372{
373	vfree(image->arch.elf_headers);
374	image->arch.elf_headers = NULL;
375
376	if (!image->fops || !image->fops->load)
377		return ERR_PTR(-ENOEXEC);
378
379	return image->fops->load(image, image->kernel_buf,
380				 image->kernel_buf_len, image->initrd_buf,
381				 image->initrd_buf_len, image->cmdline_buf,
382				 image->cmdline_buf_len);
383}
384
385/*
386 * Apply purgatory relocations.
387 *
388 * @pi:		Purgatory to be relocated.
389 * @section:	Section relocations applying to.
390 * @relsec:	Section containing RELAs.
391 * @symtabsec:	Corresponding symtab.
392 *
393 * TODO: Some of the code belongs to generic code. Move that in kexec.c.
394 */
395int arch_kexec_apply_relocations_add(struct purgatory_info *pi,
396				     Elf_Shdr *section, const Elf_Shdr *relsec,
397				     const Elf_Shdr *symtabsec)
398{
399	unsigned int i;
400	Elf64_Rela *rel;
401	Elf64_Sym *sym;
402	void *location;
403	unsigned long address, sec_base, value;
404	const char *strtab, *name, *shstrtab;
405	const Elf_Shdr *sechdrs;
406
407	/* String & section header string table */
408	sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
409	strtab = (char *)pi->ehdr + sechdrs[symtabsec->sh_link].sh_offset;
410	shstrtab = (char *)pi->ehdr + sechdrs[pi->ehdr->e_shstrndx].sh_offset;
411
412	rel = (void *)pi->ehdr + relsec->sh_offset;
413
414	pr_debug("Applying relocate section %s to %u\n",
415		 shstrtab + relsec->sh_name, relsec->sh_info);
416
417	for (i = 0; i < relsec->sh_size / sizeof(*rel); i++) {
418
419		/*
420		 * rel[i].r_offset contains byte offset from beginning
421		 * of section to the storage unit affected.
422		 *
423		 * This is location to update. This is temporary buffer
424		 * where section is currently loaded. This will finally be
425		 * loaded to a different address later, pointed to by
426		 * ->sh_addr. kexec takes care of moving it
427		 *  (kexec_load_segment()).
428		 */
429		location = pi->purgatory_buf;
430		location += section->sh_offset;
431		location += rel[i].r_offset;
432
433		/* Final address of the location */
434		address = section->sh_addr + rel[i].r_offset;
435
436		/*
437		 * rel[i].r_info contains information about symbol table index
438		 * w.r.t which relocation must be made and type of relocation
439		 * to apply. ELF64_R_SYM() and ELF64_R_TYPE() macros get
440		 * these respectively.
441		 */
442		sym = (void *)pi->ehdr + symtabsec->sh_offset;
443		sym += ELF64_R_SYM(rel[i].r_info);
444
445		if (sym->st_name)
446			name = strtab + sym->st_name;
447		else
448			name = shstrtab + sechdrs[sym->st_shndx].sh_name;
449
450		pr_debug("Symbol: %s info: %02x shndx: %02x value=%llx size: %llx\n",
451			 name, sym->st_info, sym->st_shndx, sym->st_value,
452			 sym->st_size);
453
454		if (sym->st_shndx == SHN_UNDEF) {
455			pr_err("Undefined symbol: %s\n", name);
456			return -ENOEXEC;
457		}
458
459		if (sym->st_shndx == SHN_COMMON) {
460			pr_err("symbol '%s' in common section\n", name);
461			return -ENOEXEC;
462		}
463
464		if (sym->st_shndx == SHN_ABS)
465			sec_base = 0;
466		else if (sym->st_shndx >= pi->ehdr->e_shnum) {
467			pr_err("Invalid section %d for symbol %s\n",
468			       sym->st_shndx, name);
469			return -ENOEXEC;
470		} else
471			sec_base = pi->sechdrs[sym->st_shndx].sh_addr;
472
473		value = sym->st_value;
474		value += sec_base;
475		value += rel[i].r_addend;
476
477		switch (ELF64_R_TYPE(rel[i].r_info)) {
478		case R_X86_64_NONE:
479			break;
480		case R_X86_64_64:
481			*(u64 *)location = value;
482			break;
483		case R_X86_64_32:
484			*(u32 *)location = value;
485			if (value != *(u32 *)location)
486				goto overflow;
487			break;
488		case R_X86_64_32S:
489			*(s32 *)location = value;
490			if ((s64)value != *(s32 *)location)
491				goto overflow;
492			break;
493		case R_X86_64_PC32:
494		case R_X86_64_PLT32:
495			value -= (u64)address;
496			*(u32 *)location = value;
497			break;
498		default:
499			pr_err("Unknown rela relocation: %llu\n",
500			       ELF64_R_TYPE(rel[i].r_info));
501			return -ENOEXEC;
502		}
503	}
504	return 0;
505
506overflow:
507	pr_err("Overflow in relocation type %d value 0x%lx\n",
508	       (int)ELF64_R_TYPE(rel[i].r_info), value);
509	return -ENOEXEC;
510}
511#endif /* CONFIG_KEXEC_FILE */
512
513static int
514kexec_mark_range(unsigned long start, unsigned long end, bool protect)
515{
516	struct page *page;
517	unsigned int nr_pages;
518
519	/*
520	 * For physical range: [start, end]. We must skip the unassigned
521	 * crashk resource with zero-valued "end" member.
522	 */
523	if (!end || start > end)
524		return 0;
525
526	page = pfn_to_page(start >> PAGE_SHIFT);
527	nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
528	if (protect)
529		return set_pages_ro(page, nr_pages);
530	else
531		return set_pages_rw(page, nr_pages);
532}
533
534static void kexec_mark_crashkres(bool protect)
535{
536	unsigned long control;
537
538	kexec_mark_range(crashk_low_res.start, crashk_low_res.end, protect);
539
540	/* Don't touch the control code page used in crash_kexec().*/
541	control = PFN_PHYS(page_to_pfn(kexec_crash_image->control_code_page));
542	/* Control code page is located in the 2nd page. */
543	kexec_mark_range(crashk_res.start, control + PAGE_SIZE - 1, protect);
544	control += KEXEC_CONTROL_PAGE_SIZE;
545	kexec_mark_range(control, crashk_res.end, protect);
546}
547
548void arch_kexec_protect_crashkres(void)
549{
550	kexec_mark_crashkres(true);
551}
552
553void arch_kexec_unprotect_crashkres(void)
554{
555	kexec_mark_crashkres(false);
556}
557
558int arch_kexec_post_alloc_pages(void *vaddr, unsigned int pages, gfp_t gfp)
559{
560	/*
561	 * If SME is active we need to be sure that kexec pages are
562	 * not encrypted because when we boot to the new kernel the
563	 * pages won't be accessed encrypted (initially).
564	 */
565	return set_memory_decrypted((unsigned long)vaddr, pages);
566}
567
568void arch_kexec_pre_free_pages(void *vaddr, unsigned int pages)
569{
570	/*
571	 * If SME is active we need to reset the pages back to being
572	 * an encrypted mapping before freeing them.
573	 */
574	set_memory_encrypted((unsigned long)vaddr, pages);
575}
v3.5.6
  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#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
 
 19#include <asm/pgtable.h>
 20#include <asm/tlbflush.h>
 21#include <asm/mmu_context.h>
 
 22#include <asm/debugreg.h>
 23
 24static int init_one_level2_page(struct kimage *image, pgd_t *pgd,
 25				unsigned long addr)
 26{
 27	pud_t *pud;
 28	pmd_t *pmd;
 29	struct page *page;
 30	int result = -ENOMEM;
 31
 32	addr &= PMD_MASK;
 33	pgd += pgd_index(addr);
 34	if (!pgd_present(*pgd)) {
 35		page = kimage_alloc_control_pages(image, 0);
 36		if (!page)
 37			goto out;
 38		pud = (pud_t *)page_address(page);
 39		clear_page(pud);
 40		set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE));
 41	}
 42	pud = pud_offset(pgd, addr);
 43	if (!pud_present(*pud)) {
 44		page = kimage_alloc_control_pages(image, 0);
 45		if (!page)
 46			goto out;
 47		pmd = (pmd_t *)page_address(page);
 48		clear_page(pmd);
 49		set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
 50	}
 51	pmd = pmd_offset(pud, addr);
 52	if (!pmd_present(*pmd))
 53		set_pmd(pmd, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC));
 54	result = 0;
 55out:
 56	return result;
 57}
 58
 59static void init_level2_page(pmd_t *level2p, unsigned long addr)
 60{
 61	unsigned long end_addr;
 62
 63	addr &= PAGE_MASK;
 64	end_addr = addr + PUD_SIZE;
 65	while (addr < end_addr) {
 66		set_pmd(level2p++, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC));
 67		addr += PMD_SIZE;
 68	}
 69}
 70
 71static int init_level3_page(struct kimage *image, pud_t *level3p,
 72				unsigned long addr, unsigned long last_addr)
 73{
 74	unsigned long end_addr;
 75	int result;
 76
 77	result = 0;
 78	addr &= PAGE_MASK;
 79	end_addr = addr + PGDIR_SIZE;
 80	while ((addr < last_addr) && (addr < end_addr)) {
 81		struct page *page;
 82		pmd_t *level2p;
 83
 84		page = kimage_alloc_control_pages(image, 0);
 85		if (!page) {
 86			result = -ENOMEM;
 87			goto out;
 88		}
 89		level2p = (pmd_t *)page_address(page);
 90		init_level2_page(level2p, addr);
 91		set_pud(level3p++, __pud(__pa(level2p) | _KERNPG_TABLE));
 92		addr += PUD_SIZE;
 93	}
 94	/* clear the unused entries */
 95	while (addr < end_addr) {
 96		pud_clear(level3p++);
 97		addr += PUD_SIZE;
 98	}
 99out:
100	return result;
101}
102
103
104static int init_level4_page(struct kimage *image, pgd_t *level4p,
105				unsigned long addr, unsigned long last_addr)
106{
107	unsigned long end_addr;
108	int result;
109
110	result = 0;
111	addr &= PAGE_MASK;
112	end_addr = addr + (PTRS_PER_PGD * PGDIR_SIZE);
113	while ((addr < last_addr) && (addr < end_addr)) {
114		struct page *page;
115		pud_t *level3p;
116
117		page = kimage_alloc_control_pages(image, 0);
118		if (!page) {
119			result = -ENOMEM;
120			goto out;
121		}
122		level3p = (pud_t *)page_address(page);
123		result = init_level3_page(image, level3p, addr, last_addr);
124		if (result)
125			goto out;
126		set_pgd(level4p++, __pgd(__pa(level3p) | _KERNPG_TABLE));
127		addr += PGDIR_SIZE;
128	}
129	/* clear the unused entries */
130	while (addr < end_addr) {
131		pgd_clear(level4p++);
132		addr += PGDIR_SIZE;
133	}
134out:
135	return result;
136}
137
138static void free_transition_pgtable(struct kimage *image)
139{
 
 
140	free_page((unsigned long)image->arch.pud);
 
141	free_page((unsigned long)image->arch.pmd);
 
142	free_page((unsigned long)image->arch.pte);
 
143}
144
145static int init_transition_pgtable(struct kimage *image, pgd_t *pgd)
146{
 
147	pud_t *pud;
148	pmd_t *pmd;
149	pte_t *pte;
150	unsigned long vaddr, paddr;
151	int result = -ENOMEM;
152
153	vaddr = (unsigned long)relocate_kernel;
154	paddr = __pa(page_address(image->control_code_page)+PAGE_SIZE);
155	pgd += pgd_index(vaddr);
156	if (!pgd_present(*pgd)) {
 
 
 
 
 
 
 
 
157		pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
158		if (!pud)
159			goto err;
160		image->arch.pud = pud;
161		set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE));
162	}
163	pud = pud_offset(pgd, vaddr);
164	if (!pud_present(*pud)) {
165		pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
166		if (!pmd)
167			goto err;
168		image->arch.pmd = pmd;
169		set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
170	}
171	pmd = pmd_offset(pud, vaddr);
172	if (!pmd_present(*pmd)) {
173		pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
174		if (!pte)
175			goto err;
176		image->arch.pte = pte;
177		set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
178	}
179	pte = pte_offset_kernel(pmd, vaddr);
180	set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC));
181	return 0;
182err:
183	free_transition_pgtable(image);
184	return result;
185}
186
 
 
 
 
 
 
 
 
 
 
 
 
 
 
187
188static int init_pgtable(struct kimage *image, unsigned long start_pgtable)
189{
 
 
 
 
 
 
 
190	pgd_t *level4p;
191	int result;
 
 
192	level4p = (pgd_t *)__va(start_pgtable);
193	result = init_level4_page(image, level4p, 0, max_pfn << PAGE_SHIFT);
194	if (result)
195		return result;
 
 
 
 
 
 
 
 
 
 
 
 
196	/*
197	 * image->start may be outside 0 ~ max_pfn, for example when
198	 * jump back to original kernel from kexeced kernel
 
 
199	 */
200	result = init_one_level2_page(image, level4p, image->start);
201	if (result)
202		return result;
 
 
 
 
 
 
 
 
203	return init_transition_pgtable(image, level4p);
204}
205
206static void set_idt(void *newidt, u16 limit)
207{
208	struct desc_ptr curidt;
209
210	/* x86-64 supports unaliged loads & stores */
211	curidt.size    = limit;
212	curidt.address = (unsigned long)newidt;
213
214	__asm__ __volatile__ (
215		"lidtq %0\n"
216		: : "m" (curidt)
217		);
218};
219
220
221static void set_gdt(void *newgdt, u16 limit)
222{
223	struct desc_ptr curgdt;
224
225	/* x86-64 supports unaligned loads & stores */
226	curgdt.size    = limit;
227	curgdt.address = (unsigned long)newgdt;
228
229	__asm__ __volatile__ (
230		"lgdtq %0\n"
231		: : "m" (curgdt)
232		);
233};
234
235static void load_segments(void)
236{
237	__asm__ __volatile__ (
238		"\tmovl %0,%%ds\n"
239		"\tmovl %0,%%es\n"
240		"\tmovl %0,%%ss\n"
241		"\tmovl %0,%%fs\n"
242		"\tmovl %0,%%gs\n"
243		: : "a" (__KERNEL_DS) : "memory"
244		);
245}
246
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
247int machine_kexec_prepare(struct kimage *image)
248{
249	unsigned long start_pgtable;
250	int result;
251
252	/* Calculate the offsets */
253	start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
254
255	/* Setup the identity mapped 64bit page table */
256	result = init_pgtable(image, start_pgtable);
257	if (result)
258		return result;
259
 
 
 
 
 
260	return 0;
261}
262
263void machine_kexec_cleanup(struct kimage *image)
264{
265	free_transition_pgtable(image);
266}
267
268/*
269 * Do not allocate memory (or fail in any way) in machine_kexec().
270 * We are past the point of no return, committed to rebooting now.
271 */
272void machine_kexec(struct kimage *image)
273{
274	unsigned long page_list[PAGES_NR];
275	void *control_page;
276	int save_ftrace_enabled;
277
278#ifdef CONFIG_KEXEC_JUMP
279	if (image->preserve_context)
280		save_processor_state();
281#endif
282
283	save_ftrace_enabled = __ftrace_enabled_save();
284
285	/* Interrupts aren't acceptable while we reboot */
286	local_irq_disable();
287	hw_breakpoint_disable();
288
289	if (image->preserve_context) {
290#ifdef CONFIG_X86_IO_APIC
291		/*
292		 * We need to put APICs in legacy mode so that we can
293		 * get timer interrupts in second kernel. kexec/kdump
294		 * paths already have calls to disable_IO_APIC() in
295		 * one form or other. kexec jump path also need
296		 * one.
297		 */
298		disable_IO_APIC();
 
299#endif
300	}
301
302	control_page = page_address(image->control_code_page) + PAGE_SIZE;
303	memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
304
305	page_list[PA_CONTROL_PAGE] = virt_to_phys(control_page);
306	page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
307	page_list[PA_TABLE_PAGE] =
308	  (unsigned long)__pa(page_address(image->control_code_page));
309
310	if (image->type == KEXEC_TYPE_DEFAULT)
311		page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
312						<< PAGE_SHIFT);
313
314	/*
315	 * The segment registers are funny things, they have both a
316	 * visible and an invisible part.  Whenever the visible part is
317	 * set to a specific selector, the invisible part is loaded
318	 * with from a table in memory.  At no other time is the
319	 * descriptor table in memory accessed.
320	 *
321	 * I take advantage of this here by force loading the
322	 * segments, before I zap the gdt with an invalid value.
323	 */
324	load_segments();
325	/*
326	 * The gdt & idt are now invalid.
327	 * If you want to load them you must set up your own idt & gdt.
328	 */
329	set_gdt(phys_to_virt(0), 0);
330	set_idt(phys_to_virt(0), 0);
331
332	/* now call it */
333	image->start = relocate_kernel((unsigned long)image->head,
334				       (unsigned long)page_list,
335				       image->start,
336				       image->preserve_context);
 
337
338#ifdef CONFIG_KEXEC_JUMP
339	if (image->preserve_context)
340		restore_processor_state();
341#endif
342
343	__ftrace_enabled_restore(save_ftrace_enabled);
344}
345
346void arch_crash_save_vmcoreinfo(void)
347{
348	VMCOREINFO_SYMBOL(phys_base);
349	VMCOREINFO_SYMBOL(init_level4_pgt);
 
350
351#ifdef CONFIG_NUMA
352	VMCOREINFO_SYMBOL(node_data);
353	VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);
354#endif
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
355}
356