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/cpufeature.h>
24 #include <asm/desc.h>
25 #include <asm/cacheflush.h>
26 #include <asm/debugreg.h>
27 
28 static void set_idt(void *newidt, __u16 limit)
29 {
30 	struct desc_ptr curidt;
31 
32 	/* ia32 supports unaliged loads & stores */
33 	curidt.size    = limit;
34 	curidt.address = (unsigned long)newidt;
35 
36 	load_idt(&curidt);
37 }
38 
39 
40 static void set_gdt(void *newgdt, __u16 limit)
41 {
42 	struct desc_ptr curgdt;
43 
44 	/* ia32 supports unaligned loads & stores */
45 	curgdt.size    = limit;
46 	curgdt.address = (unsigned long)newgdt;
47 
48 	load_gdt(&curgdt);
49 }
50 
51 static void load_segments(void)
52 {
53 #define __STR(X) #X
54 #define STR(X) __STR(X)
55 
56 	__asm__ __volatile__ (
57 		"\tljmp $"STR(__KERNEL_CS)",$1f\n"
58 		"\t1:\n"
59 		"\tmovl $"STR(__KERNEL_DS)",%%eax\n"
60 		"\tmovl %%eax,%%ds\n"
61 		"\tmovl %%eax,%%es\n"
62 		"\tmovl %%eax,%%fs\n"
63 		"\tmovl %%eax,%%gs\n"
64 		"\tmovl %%eax,%%ss\n"
65 		: : : "eax", "memory");
66 #undef STR
67 #undef __STR
68 }
69 
70 static void machine_kexec_free_page_tables(struct kimage *image)
71 {
72 	free_page((unsigned long)image->arch.pgd);
73 #ifdef CONFIG_X86_PAE
74 	free_page((unsigned long)image->arch.pmd0);
75 	free_page((unsigned long)image->arch.pmd1);
76 #endif
77 	free_page((unsigned long)image->arch.pte0);
78 	free_page((unsigned long)image->arch.pte1);
79 }
80 
81 static int machine_kexec_alloc_page_tables(struct kimage *image)
82 {
83 	image->arch.pgd = (pgd_t *)get_zeroed_page(GFP_KERNEL);
84 #ifdef CONFIG_X86_PAE
85 	image->arch.pmd0 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
86 	image->arch.pmd1 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
87 #endif
88 	image->arch.pte0 = (pte_t *)get_zeroed_page(GFP_KERNEL);
89 	image->arch.pte1 = (pte_t *)get_zeroed_page(GFP_KERNEL);
90 	if (!image->arch.pgd ||
91 #ifdef CONFIG_X86_PAE
92 	    !image->arch.pmd0 || !image->arch.pmd1 ||
93 #endif
94 	    !image->arch.pte0 || !image->arch.pte1) {
95 		machine_kexec_free_page_tables(image);
96 		return -ENOMEM;
97 	}
98 	return 0;
99 }
100 
101 static void machine_kexec_page_table_set_one(
102 	pgd_t *pgd, pmd_t *pmd, pte_t *pte,
103 	unsigned long vaddr, unsigned long paddr)
104 {
105 	pud_t *pud;
106 
107 	pgd += pgd_index(vaddr);
108 #ifdef CONFIG_X86_PAE
109 	if (!(pgd_val(*pgd) & _PAGE_PRESENT))
110 		set_pgd(pgd, __pgd(__pa(pmd) | _PAGE_PRESENT));
111 #endif
112 	pud = pud_offset(pgd, vaddr);
113 	pmd = pmd_offset(pud, vaddr);
114 	if (!(pmd_val(*pmd) & _PAGE_PRESENT))
115 		set_pmd(pmd, __pmd(__pa(pte) | _PAGE_TABLE));
116 	pte = pte_offset_kernel(pmd, vaddr);
117 	set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC));
118 }
119 
120 static void machine_kexec_prepare_page_tables(struct kimage *image)
121 {
122 	void *control_page;
123 	pmd_t *pmd = NULL;
124 
125 	control_page = page_address(image->control_code_page);
126 #ifdef CONFIG_X86_PAE
127 	pmd = image->arch.pmd0;
128 #endif
129 	machine_kexec_page_table_set_one(
130 		image->arch.pgd, pmd, image->arch.pte0,
131 		(unsigned long)control_page, __pa(control_page));
132 #ifdef CONFIG_X86_PAE
133 	pmd = image->arch.pmd1;
134 #endif
135 	machine_kexec_page_table_set_one(
136 		image->arch.pgd, pmd, image->arch.pte1,
137 		__pa(control_page), __pa(control_page));
138 }
139 
140 /*
141  * A architecture hook called to validate the
142  * proposed image and prepare the control pages
143  * as needed.  The pages for KEXEC_CONTROL_PAGE_SIZE
144  * have been allocated, but the segments have yet
145  * been copied into the kernel.
146  *
147  * Do what every setup is needed on image and the
148  * reboot code buffer to allow us to avoid allocations
149  * later.
150  *
151  * - Make control page executable.
152  * - Allocate page tables
153  * - Setup page tables
154  */
155 int machine_kexec_prepare(struct kimage *image)
156 {
157 	int error;
158 
159 	set_pages_x(image->control_code_page, 1);
160 	error = machine_kexec_alloc_page_tables(image);
161 	if (error)
162 		return error;
163 	machine_kexec_prepare_page_tables(image);
164 	return 0;
165 }
166 
167 /*
168  * Undo anything leftover by machine_kexec_prepare
169  * when an image is freed.
170  */
171 void machine_kexec_cleanup(struct kimage *image)
172 {
173 	set_pages_nx(image->control_code_page, 1);
174 	machine_kexec_free_page_tables(image);
175 }
176 
177 /*
178  * Do not allocate memory (or fail in any way) in machine_kexec().
179  * We are past the point of no return, committed to rebooting now.
180  */
181 void machine_kexec(struct kimage *image)
182 {
183 	unsigned long page_list[PAGES_NR];
184 	void *control_page;
185 	int save_ftrace_enabled;
186 	asmlinkage unsigned long
187 		(*relocate_kernel_ptr)(unsigned long indirection_page,
188 				       unsigned long control_page,
189 				       unsigned long start_address,
190 				       unsigned int has_pae,
191 				       unsigned int preserve_context);
192 
193 #ifdef CONFIG_KEXEC_JUMP
194 	if (image->preserve_context)
195 		save_processor_state();
196 #endif
197 
198 	save_ftrace_enabled = __ftrace_enabled_save();
199 
200 	/* Interrupts aren't acceptable while we reboot */
201 	local_irq_disable();
202 	hw_breakpoint_disable();
203 
204 	if (image->preserve_context) {
205 #ifdef CONFIG_X86_IO_APIC
206 		/*
207 		 * We need to put APICs in legacy mode so that we can
208 		 * get timer interrupts in second kernel. kexec/kdump
209 		 * paths already have calls to disable_IO_APIC() in
210 		 * one form or other. kexec jump path also need
211 		 * one.
212 		 */
213 		disable_IO_APIC();
214 #endif
215 	}
216 
217 	control_page = page_address(image->control_code_page);
218 	memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
219 
220 	relocate_kernel_ptr = control_page;
221 	page_list[PA_CONTROL_PAGE] = __pa(control_page);
222 	page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
223 	page_list[PA_PGD] = __pa(image->arch.pgd);
224 
225 	if (image->type == KEXEC_TYPE_DEFAULT)
226 		page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
227 						<< PAGE_SHIFT);
228 
229 	/*
230 	 * The segment registers are funny things, they have both a
231 	 * visible and an invisible part.  Whenever the visible part is
232 	 * set to a specific selector, the invisible part is loaded
233 	 * with from a table in memory.  At no other time is the
234 	 * descriptor table in memory accessed.
235 	 *
236 	 * I take advantage of this here by force loading the
237 	 * segments, before I zap the gdt with an invalid value.
238 	 */
239 	load_segments();
240 	/*
241 	 * The gdt & idt are now invalid.
242 	 * If you want to load them you must set up your own idt & gdt.
243 	 */
244 	set_gdt(phys_to_virt(0), 0);
245 	set_idt(phys_to_virt(0), 0);
246 
247 	/* now call it */
248 	image->start = relocate_kernel_ptr((unsigned long)image->head,
249 					   (unsigned long)page_list,
250 					   image->start,
251 					   boot_cpu_has(X86_FEATURE_PAE),
252 					   image->preserve_context);
253 
254 #ifdef CONFIG_KEXEC_JUMP
255 	if (image->preserve_context)
256 		restore_processor_state();
257 #endif
258 
259 	__ftrace_enabled_restore(save_ftrace_enabled);
260 }
261 
262 void arch_crash_save_vmcoreinfo(void)
263 {
264 #ifdef CONFIG_NUMA
265 	VMCOREINFO_SYMBOL(node_data);
266 	VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);
267 #endif
268 #ifdef CONFIG_X86_PAE
269 	VMCOREINFO_CONFIG(X86_PAE);
270 #endif
271 }
272 
273