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/delay.h>
12 #include <linux/numa.h>
13 #include <linux/ftrace.h>
14 #include <linux/suspend.h>
15 #include <linux/gfp.h>
16 #include <linux/io.h>
17 
18 #include <asm/pgtable.h>
19 #include <asm/pgalloc.h>
20 #include <asm/tlbflush.h>
21 #include <asm/mmu_context.h>
22 #include <asm/apic.h>
23 #include <asm/io_apic.h>
24 #include <asm/cpufeature.h>
25 #include <asm/desc.h>
26 #include <asm/set_memory.h>
27 #include <asm/debugreg.h>
28 
29 static void set_gdt(void *newgdt, __u16 limit)
30 {
31 	struct desc_ptr curgdt;
32 
33 	/* ia32 supports unaligned loads & stores */
34 	curgdt.size    = limit;
35 	curgdt.address = (unsigned long)newgdt;
36 
37 	load_gdt(&curgdt);
38 }
39 
40 static void load_segments(void)
41 {
42 #define __STR(X) #X
43 #define STR(X) __STR(X)
44 
45 	__asm__ __volatile__ (
46 		"\tljmp $"STR(__KERNEL_CS)",$1f\n"
47 		"\t1:\n"
48 		"\tmovl $"STR(__KERNEL_DS)",%%eax\n"
49 		"\tmovl %%eax,%%ds\n"
50 		"\tmovl %%eax,%%es\n"
51 		"\tmovl %%eax,%%ss\n"
52 		: : : "eax", "memory");
53 #undef STR
54 #undef __STR
55 }
56 
57 static void machine_kexec_free_page_tables(struct kimage *image)
58 {
59 	free_pages((unsigned long)image->arch.pgd, PGD_ALLOCATION_ORDER);
60 	image->arch.pgd = NULL;
61 #ifdef CONFIG_X86_PAE
62 	free_page((unsigned long)image->arch.pmd0);
63 	image->arch.pmd0 = NULL;
64 	free_page((unsigned long)image->arch.pmd1);
65 	image->arch.pmd1 = NULL;
66 #endif
67 	free_page((unsigned long)image->arch.pte0);
68 	image->arch.pte0 = NULL;
69 	free_page((unsigned long)image->arch.pte1);
70 	image->arch.pte1 = NULL;
71 }
72 
73 static int machine_kexec_alloc_page_tables(struct kimage *image)
74 {
75 	image->arch.pgd = (pgd_t *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
76 						    PGD_ALLOCATION_ORDER);
77 #ifdef CONFIG_X86_PAE
78 	image->arch.pmd0 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
79 	image->arch.pmd1 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
80 #endif
81 	image->arch.pte0 = (pte_t *)get_zeroed_page(GFP_KERNEL);
82 	image->arch.pte1 = (pte_t *)get_zeroed_page(GFP_KERNEL);
83 	if (!image->arch.pgd ||
84 #ifdef CONFIG_X86_PAE
85 	    !image->arch.pmd0 || !image->arch.pmd1 ||
86 #endif
87 	    !image->arch.pte0 || !image->arch.pte1) {
88 		return -ENOMEM;
89 	}
90 	return 0;
91 }
92 
93 static void machine_kexec_page_table_set_one(
94 	pgd_t *pgd, pmd_t *pmd, pte_t *pte,
95 	unsigned long vaddr, unsigned long paddr)
96 {
97 	p4d_t *p4d;
98 	pud_t *pud;
99 
100 	pgd += pgd_index(vaddr);
101 #ifdef CONFIG_X86_PAE
102 	if (!(pgd_val(*pgd) & _PAGE_PRESENT))
103 		set_pgd(pgd, __pgd(__pa(pmd) | _PAGE_PRESENT));
104 #endif
105 	p4d = p4d_offset(pgd, vaddr);
106 	pud = pud_offset(p4d, vaddr);
107 	pmd = pmd_offset(pud, vaddr);
108 	if (!(pmd_val(*pmd) & _PAGE_PRESENT))
109 		set_pmd(pmd, __pmd(__pa(pte) | _PAGE_TABLE));
110 	pte = pte_offset_kernel(pmd, vaddr);
111 	set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC));
112 }
113 
114 static void machine_kexec_prepare_page_tables(struct kimage *image)
115 {
116 	void *control_page;
117 	pmd_t *pmd = NULL;
118 
119 	control_page = page_address(image->control_code_page);
120 #ifdef CONFIG_X86_PAE
121 	pmd = image->arch.pmd0;
122 #endif
123 	machine_kexec_page_table_set_one(
124 		image->arch.pgd, pmd, image->arch.pte0,
125 		(unsigned long)control_page, __pa(control_page));
126 #ifdef CONFIG_X86_PAE
127 	pmd = image->arch.pmd1;
128 #endif
129 	machine_kexec_page_table_set_one(
130 		image->arch.pgd, pmd, image->arch.pte1,
131 		__pa(control_page), __pa(control_page));
132 }
133 
134 /*
135  * A architecture hook called to validate the
136  * proposed image and prepare the control pages
137  * as needed.  The pages for KEXEC_CONTROL_PAGE_SIZE
138  * have been allocated, but the segments have yet
139  * been copied into the kernel.
140  *
141  * Do what every setup is needed on image and the
142  * reboot code buffer to allow us to avoid allocations
143  * later.
144  *
145  * - Make control page executable.
146  * - Allocate page tables
147  * - Setup page tables
148  */
149 int machine_kexec_prepare(struct kimage *image)
150 {
151 	int error;
152 
153 	set_pages_x(image->control_code_page, 1);
154 	error = machine_kexec_alloc_page_tables(image);
155 	if (error)
156 		return error;
157 	machine_kexec_prepare_page_tables(image);
158 	return 0;
159 }
160 
161 /*
162  * Undo anything leftover by machine_kexec_prepare
163  * when an image is freed.
164  */
165 void machine_kexec_cleanup(struct kimage *image)
166 {
167 	set_pages_nx(image->control_code_page, 1);
168 	machine_kexec_free_page_tables(image);
169 }
170 
171 /*
172  * Do not allocate memory (or fail in any way) in machine_kexec().
173  * We are past the point of no return, committed to rebooting now.
174  */
175 void machine_kexec(struct kimage *image)
176 {
177 	unsigned long page_list[PAGES_NR];
178 	void *control_page;
179 	int save_ftrace_enabled;
180 	asmlinkage unsigned long
181 		(*relocate_kernel_ptr)(unsigned long indirection_page,
182 				       unsigned long control_page,
183 				       unsigned long start_address,
184 				       unsigned int has_pae,
185 				       unsigned int preserve_context);
186 
187 #ifdef CONFIG_KEXEC_JUMP
188 	if (image->preserve_context)
189 		save_processor_state();
190 #endif
191 
192 	save_ftrace_enabled = __ftrace_enabled_save();
193 
194 	/* Interrupts aren't acceptable while we reboot */
195 	local_irq_disable();
196 	hw_breakpoint_disable();
197 
198 	if (image->preserve_context) {
199 #ifdef CONFIG_X86_IO_APIC
200 		/*
201 		 * We need to put APICs in legacy mode so that we can
202 		 * get timer interrupts in second kernel. kexec/kdump
203 		 * paths already have calls to restore_boot_irq_mode()
204 		 * in one form or other. kexec jump path also need one.
205 		 */
206 		clear_IO_APIC();
207 		restore_boot_irq_mode();
208 #endif
209 	}
210 
211 	control_page = page_address(image->control_code_page);
212 	memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
213 
214 	relocate_kernel_ptr = control_page;
215 	page_list[PA_CONTROL_PAGE] = __pa(control_page);
216 	page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
217 	page_list[PA_PGD] = __pa(image->arch.pgd);
218 
219 	if (image->type == KEXEC_TYPE_DEFAULT)
220 		page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
221 						<< PAGE_SHIFT);
222 
223 	/*
224 	 * The segment registers are funny things, they have both a
225 	 * visible and an invisible part.  Whenever the visible part is
226 	 * set to a specific selector, the invisible part is loaded
227 	 * with from a table in memory.  At no other time is the
228 	 * descriptor table in memory accessed.
229 	 *
230 	 * I take advantage of this here by force loading the
231 	 * segments, before I zap the gdt with an invalid value.
232 	 */
233 	load_segments();
234 	/*
235 	 * The gdt & idt are now invalid.
236 	 * If you want to load them you must set up your own idt & gdt.
237 	 */
238 	idt_invalidate(phys_to_virt(0));
239 	set_gdt(phys_to_virt(0), 0);
240 
241 	/* now call it */
242 	image->start = relocate_kernel_ptr((unsigned long)image->head,
243 					   (unsigned long)page_list,
244 					   image->start,
245 					   boot_cpu_has(X86_FEATURE_PAE),
246 					   image->preserve_context);
247 
248 #ifdef CONFIG_KEXEC_JUMP
249 	if (image->preserve_context)
250 		restore_processor_state();
251 #endif
252 
253 	__ftrace_enabled_restore(save_ftrace_enabled);
254 }
255 
256 void arch_crash_save_vmcoreinfo(void)
257 {
258 #ifdef CONFIG_NUMA
259 	VMCOREINFO_SYMBOL(node_data);
260 	VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);
261 #endif
262 #ifdef CONFIG_X86_PAE
263 	VMCOREINFO_CONFIG(X86_PAE);
264 #endif
265 }
266 
267