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_page((unsigned long)image->arch.pgd);
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_zeroed_page(GFP_KERNEL);
76 #ifdef CONFIG_X86_PAE
77 	image->arch.pmd0 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
78 	image->arch.pmd1 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
79 #endif
80 	image->arch.pte0 = (pte_t *)get_zeroed_page(GFP_KERNEL);
81 	image->arch.pte1 = (pte_t *)get_zeroed_page(GFP_KERNEL);
82 	if (!image->arch.pgd ||
83 #ifdef CONFIG_X86_PAE
84 	    !image->arch.pmd0 || !image->arch.pmd1 ||
85 #endif
86 	    !image->arch.pte0 || !image->arch.pte1) {
87 		return -ENOMEM;
88 	}
89 	return 0;
90 }
91 
92 static void machine_kexec_page_table_set_one(
93 	pgd_t *pgd, pmd_t *pmd, pte_t *pte,
94 	unsigned long vaddr, unsigned long paddr)
95 {
96 	p4d_t *p4d;
97 	pud_t *pud;
98 
99 	pgd += pgd_index(vaddr);
100 #ifdef CONFIG_X86_PAE
101 	if (!(pgd_val(*pgd) & _PAGE_PRESENT))
102 		set_pgd(pgd, __pgd(__pa(pmd) | _PAGE_PRESENT));
103 #endif
104 	p4d = p4d_offset(pgd, vaddr);
105 	pud = pud_offset(p4d, vaddr);
106 	pmd = pmd_offset(pud, vaddr);
107 	if (!(pmd_val(*pmd) & _PAGE_PRESENT))
108 		set_pmd(pmd, __pmd(__pa(pte) | _PAGE_TABLE));
109 	pte = pte_offset_kernel(pmd, vaddr);
110 	set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC));
111 }
112 
113 static void machine_kexec_prepare_page_tables(struct kimage *image)
114 {
115 	void *control_page;
116 	pmd_t *pmd = NULL;
117 
118 	control_page = page_address(image->control_code_page);
119 #ifdef CONFIG_X86_PAE
120 	pmd = image->arch.pmd0;
121 #endif
122 	machine_kexec_page_table_set_one(
123 		image->arch.pgd, pmd, image->arch.pte0,
124 		(unsigned long)control_page, __pa(control_page));
125 #ifdef CONFIG_X86_PAE
126 	pmd = image->arch.pmd1;
127 #endif
128 	machine_kexec_page_table_set_one(
129 		image->arch.pgd, pmd, image->arch.pte1,
130 		__pa(control_page), __pa(control_page));
131 }
132 
133 /*
134  * A architecture hook called to validate the
135  * proposed image and prepare the control pages
136  * as needed.  The pages for KEXEC_CONTROL_PAGE_SIZE
137  * have been allocated, but the segments have yet
138  * been copied into the kernel.
139  *
140  * Do what every setup is needed on image and the
141  * reboot code buffer to allow us to avoid allocations
142  * later.
143  *
144  * - Make control page executable.
145  * - Allocate page tables
146  * - Setup page tables
147  */
148 int machine_kexec_prepare(struct kimage *image)
149 {
150 	int error;
151 
152 	set_pages_x(image->control_code_page, 1);
153 	error = machine_kexec_alloc_page_tables(image);
154 	if (error)
155 		return error;
156 	machine_kexec_prepare_page_tables(image);
157 	return 0;
158 }
159 
160 /*
161  * Undo anything leftover by machine_kexec_prepare
162  * when an image is freed.
163  */
164 void machine_kexec_cleanup(struct kimage *image)
165 {
166 	set_pages_nx(image->control_code_page, 1);
167 	machine_kexec_free_page_tables(image);
168 }
169 
170 /*
171  * Do not allocate memory (or fail in any way) in machine_kexec().
172  * We are past the point of no return, committed to rebooting now.
173  */
174 void machine_kexec(struct kimage *image)
175 {
176 	unsigned long page_list[PAGES_NR];
177 	void *control_page;
178 	int save_ftrace_enabled;
179 	asmlinkage unsigned long
180 		(*relocate_kernel_ptr)(unsigned long indirection_page,
181 				       unsigned long control_page,
182 				       unsigned long start_address,
183 				       unsigned int has_pae,
184 				       unsigned int preserve_context);
185 
186 #ifdef CONFIG_KEXEC_JUMP
187 	if (image->preserve_context)
188 		save_processor_state();
189 #endif
190 
191 	save_ftrace_enabled = __ftrace_enabled_save();
192 
193 	/* Interrupts aren't acceptable while we reboot */
194 	local_irq_disable();
195 	hw_breakpoint_disable();
196 
197 	if (image->preserve_context) {
198 #ifdef CONFIG_X86_IO_APIC
199 		/*
200 		 * We need to put APICs in legacy mode so that we can
201 		 * get timer interrupts in second kernel. kexec/kdump
202 		 * paths already have calls to restore_boot_irq_mode()
203 		 * in one form or other. kexec jump path also need one.
204 		 */
205 		clear_IO_APIC();
206 		restore_boot_irq_mode();
207 #endif
208 	}
209 
210 	control_page = page_address(image->control_code_page);
211 	memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
212 
213 	relocate_kernel_ptr = control_page;
214 	page_list[PA_CONTROL_PAGE] = __pa(control_page);
215 	page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
216 	page_list[PA_PGD] = __pa(image->arch.pgd);
217 
218 	if (image->type == KEXEC_TYPE_DEFAULT)
219 		page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
220 						<< PAGE_SHIFT);
221 
222 	/*
223 	 * The segment registers are funny things, they have both a
224 	 * visible and an invisible part.  Whenever the visible part is
225 	 * set to a specific selector, the invisible part is loaded
226 	 * with from a table in memory.  At no other time is the
227 	 * descriptor table in memory accessed.
228 	 *
229 	 * I take advantage of this here by force loading the
230 	 * segments, before I zap the gdt with an invalid value.
231 	 */
232 	load_segments();
233 	/*
234 	 * The gdt & idt are now invalid.
235 	 * If you want to load them you must set up your own idt & gdt.
236 	 */
237 	idt_invalidate(phys_to_virt(0));
238 	set_gdt(phys_to_virt(0), 0);
239 
240 	/* now call it */
241 	image->start = relocate_kernel_ptr((unsigned long)image->head,
242 					   (unsigned long)page_list,
243 					   image->start,
244 					   boot_cpu_has(X86_FEATURE_PAE),
245 					   image->preserve_context);
246 
247 #ifdef CONFIG_KEXEC_JUMP
248 	if (image->preserve_context)
249 		restore_processor_state();
250 #endif
251 
252 	__ftrace_enabled_restore(save_ftrace_enabled);
253 }
254 
255 void arch_crash_save_vmcoreinfo(void)
256 {
257 #ifdef CONFIG_NUMA
258 	VMCOREINFO_SYMBOL(node_data);
259 	VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);
260 #endif
261 #ifdef CONFIG_X86_PAE
262 	VMCOREINFO_CONFIG(X86_PAE);
263 #endif
264 }
265 
266