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