Commit | Line | Data |
---|---|---|
5033cba0 | 1 | /* |
835c34a1 | 2 | * handle transition of Linux booting another kernel |
5033cba0 EB |
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> | |
1a3f239d | 12 | #include <linux/init.h> |
fd59d231 | 13 | #include <linux/numa.h> |
f43fdad8 | 14 | #include <linux/ftrace.h> |
3122c331 | 15 | #include <linux/suspend.h> |
92be3d6b | 16 | #include <linux/gfp.h> |
f43fdad8 | 17 | |
5033cba0 EB |
18 | #include <asm/pgtable.h> |
19 | #include <asm/pgalloc.h> | |
20 | #include <asm/tlbflush.h> | |
21 | #include <asm/mmu_context.h> | |
22 | #include <asm/io.h> | |
23 | #include <asm/apic.h> | |
24 | #include <asm/cpufeature.h> | |
e7b47cca | 25 | #include <asm/desc.h> |
4bb0d3ec | 26 | #include <asm/system.h> |
3ab83521 | 27 | #include <asm/cacheflush.h> |
5033cba0 | 28 | |
5033cba0 EB |
29 | static void set_idt(void *newidt, __u16 limit) |
30 | { | |
6b68f01b | 31 | struct desc_ptr curidt; |
5033cba0 EB |
32 | |
33 | /* ia32 supports unaliged loads & stores */ | |
e7b47cca EB |
34 | curidt.size = limit; |
35 | curidt.address = (unsigned long)newidt; | |
5033cba0 | 36 | |
f2ab4461 | 37 | load_idt(&curidt); |
378fc6ee | 38 | } |
5033cba0 EB |
39 | |
40 | ||
41 | static void set_gdt(void *newgdt, __u16 limit) | |
42 | { | |
6b68f01b | 43 | struct desc_ptr curgdt; |
5033cba0 EB |
44 | |
45 | /* ia32 supports unaligned loads & stores */ | |
e7b47cca EB |
46 | curgdt.size = limit; |
47 | curgdt.address = (unsigned long)newgdt; | |
5033cba0 | 48 | |
f2ab4461 | 49 | load_gdt(&curgdt); |
378fc6ee | 50 | } |
5033cba0 EB |
51 | |
52 | static void load_segments(void) | |
53 | { | |
54 | #define __STR(X) #X | |
55 | #define STR(X) __STR(X) | |
56 | ||
57 | __asm__ __volatile__ ( | |
58 | "\tljmp $"STR(__KERNEL_CS)",$1f\n" | |
59 | "\t1:\n" | |
2ec5e3a8 MM |
60 | "\tmovl $"STR(__KERNEL_DS)",%%eax\n" |
61 | "\tmovl %%eax,%%ds\n" | |
62 | "\tmovl %%eax,%%es\n" | |
63 | "\tmovl %%eax,%%fs\n" | |
64 | "\tmovl %%eax,%%gs\n" | |
65 | "\tmovl %%eax,%%ss\n" | |
66 | ::: "eax", "memory"); | |
5033cba0 EB |
67 | #undef STR |
68 | #undef __STR | |
69 | } | |
70 | ||
92be3d6b HY |
71 | static void machine_kexec_free_page_tables(struct kimage *image) |
72 | { | |
73 | free_page((unsigned long)image->arch.pgd); | |
74 | #ifdef CONFIG_X86_PAE | |
75 | free_page((unsigned long)image->arch.pmd0); | |
76 | free_page((unsigned long)image->arch.pmd1); | |
77 | #endif | |
78 | free_page((unsigned long)image->arch.pte0); | |
79 | free_page((unsigned long)image->arch.pte1); | |
80 | } | |
81 | ||
82 | static int machine_kexec_alloc_page_tables(struct kimage *image) | |
83 | { | |
84 | image->arch.pgd = (pgd_t *)get_zeroed_page(GFP_KERNEL); | |
85 | #ifdef CONFIG_X86_PAE | |
86 | image->arch.pmd0 = (pmd_t *)get_zeroed_page(GFP_KERNEL); | |
87 | image->arch.pmd1 = (pmd_t *)get_zeroed_page(GFP_KERNEL); | |
88 | #endif | |
89 | image->arch.pte0 = (pte_t *)get_zeroed_page(GFP_KERNEL); | |
90 | image->arch.pte1 = (pte_t *)get_zeroed_page(GFP_KERNEL); | |
91 | if (!image->arch.pgd || | |
92 | #ifdef CONFIG_X86_PAE | |
93 | !image->arch.pmd0 || !image->arch.pmd1 || | |
94 | #endif | |
95 | !image->arch.pte0 || !image->arch.pte1) { | |
96 | machine_kexec_free_page_tables(image); | |
97 | return -ENOMEM; | |
98 | } | |
99 | return 0; | |
100 | } | |
101 | ||
9868ee63 HY |
102 | static void machine_kexec_page_table_set_one( |
103 | pgd_t *pgd, pmd_t *pmd, pte_t *pte, | |
104 | unsigned long vaddr, unsigned long paddr) | |
105 | { | |
106 | pud_t *pud; | |
107 | ||
108 | pgd += pgd_index(vaddr); | |
109 | #ifdef CONFIG_X86_PAE | |
110 | if (!(pgd_val(*pgd) & _PAGE_PRESENT)) | |
111 | set_pgd(pgd, __pgd(__pa(pmd) | _PAGE_PRESENT)); | |
112 | #endif | |
113 | pud = pud_offset(pgd, vaddr); | |
114 | pmd = pmd_offset(pud, vaddr); | |
115 | if (!(pmd_val(*pmd) & _PAGE_PRESENT)) | |
116 | set_pmd(pmd, __pmd(__pa(pte) | _PAGE_TABLE)); | |
117 | pte = pte_offset_kernel(pmd, vaddr); | |
118 | set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC)); | |
119 | } | |
120 | ||
121 | static void machine_kexec_prepare_page_tables(struct kimage *image) | |
122 | { | |
123 | void *control_page; | |
124 | pmd_t *pmd = 0; | |
125 | ||
126 | control_page = page_address(image->control_code_page); | |
127 | #ifdef CONFIG_X86_PAE | |
128 | pmd = image->arch.pmd0; | |
129 | #endif | |
130 | machine_kexec_page_table_set_one( | |
131 | image->arch.pgd, pmd, image->arch.pte0, | |
132 | (unsigned long)control_page, __pa(control_page)); | |
133 | #ifdef CONFIG_X86_PAE | |
134 | pmd = image->arch.pmd1; | |
135 | #endif | |
136 | machine_kexec_page_table_set_one( | |
137 | image->arch.pgd, pmd, image->arch.pte1, | |
138 | __pa(control_page), __pa(control_page)); | |
139 | } | |
140 | ||
5033cba0 EB |
141 | /* |
142 | * A architecture hook called to validate the | |
143 | * proposed image and prepare the control pages | |
163f6876 | 144 | * as needed. The pages for KEXEC_CONTROL_PAGE_SIZE |
5033cba0 EB |
145 | * have been allocated, but the segments have yet |
146 | * been copied into the kernel. | |
147 | * | |
148 | * Do what every setup is needed on image and the | |
149 | * reboot code buffer to allow us to avoid allocations | |
150 | * later. | |
151 | * | |
92be3d6b HY |
152 | * - Make control page executable. |
153 | * - Allocate page tables | |
9868ee63 | 154 | * - Setup page tables |
5033cba0 EB |
155 | */ |
156 | int machine_kexec_prepare(struct kimage *image) | |
157 | { | |
9868ee63 HY |
158 | int error; |
159 | ||
3ab83521 HY |
160 | if (nx_enabled) |
161 | set_pages_x(image->control_code_page, 1); | |
9868ee63 HY |
162 | error = machine_kexec_alloc_page_tables(image); |
163 | if (error) | |
164 | return error; | |
165 | machine_kexec_prepare_page_tables(image); | |
166 | return 0; | |
5033cba0 EB |
167 | } |
168 | ||
169 | /* | |
170 | * Undo anything leftover by machine_kexec_prepare | |
171 | * when an image is freed. | |
172 | */ | |
173 | void machine_kexec_cleanup(struct kimage *image) | |
174 | { | |
3ab83521 HY |
175 | if (nx_enabled) |
176 | set_pages_nx(image->control_code_page, 1); | |
92be3d6b | 177 | machine_kexec_free_page_tables(image); |
5033cba0 EB |
178 | } |
179 | ||
180 | /* | |
181 | * Do not allocate memory (or fail in any way) in machine_kexec(). | |
182 | * We are past the point of no return, committed to rebooting now. | |
183 | */ | |
3ab83521 | 184 | void machine_kexec(struct kimage *image) |
5033cba0 | 185 | { |
3566561b MD |
186 | unsigned long page_list[PAGES_NR]; |
187 | void *control_page; | |
3122c331 | 188 | int save_ftrace_enabled; |
3ab83521 HY |
189 | asmlinkage unsigned long |
190 | (*relocate_kernel_ptr)(unsigned long indirection_page, | |
191 | unsigned long control_page, | |
192 | unsigned long start_address, | |
193 | unsigned int has_pae, | |
194 | unsigned int preserve_context); | |
5033cba0 | 195 | |
3122c331 HY |
196 | #ifdef CONFIG_KEXEC_JUMP |
197 | if (kexec_image->preserve_context) | |
198 | save_processor_state(); | |
199 | #endif | |
200 | ||
201 | save_ftrace_enabled = __ftrace_enabled_save(); | |
f43fdad8 | 202 | |
5033cba0 EB |
203 | /* Interrupts aren't acceptable while we reboot */ |
204 | local_irq_disable(); | |
205 | ||
89081d17 HY |
206 | if (image->preserve_context) { |
207 | #ifdef CONFIG_X86_IO_APIC | |
208 | /* We need to put APICs in legacy mode so that we can | |
209 | * get timer interrupts in second kernel. kexec/kdump | |
210 | * paths already have calls to disable_IO_APIC() in | |
211 | * one form or other. kexec jump path also need | |
212 | * one. | |
213 | */ | |
214 | disable_IO_APIC(); | |
215 | #endif | |
216 | } | |
217 | ||
3566561b | 218 | control_page = page_address(image->control_code_page); |
fb45daa6 | 219 | memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE); |
3566561b | 220 | |
3ab83521 | 221 | relocate_kernel_ptr = control_page; |
3566561b | 222 | page_list[PA_CONTROL_PAGE] = __pa(control_page); |
3ab83521 | 223 | page_list[VA_CONTROL_PAGE] = (unsigned long)control_page; |
92be3d6b | 224 | page_list[PA_PGD] = __pa(image->arch.pgd); |
e7706fc6 KO |
225 | |
226 | if (image->type == KEXEC_TYPE_DEFAULT) | |
227 | page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page) | |
228 | << PAGE_SHIFT); | |
5033cba0 | 229 | |
2a8a3d5b EB |
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. | |
5033cba0 EB |
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 | /* The gdt & idt are now invalid. | |
241 | * If you want to load them you must set up your own idt & gdt. | |
242 | */ | |
243 | set_gdt(phys_to_virt(0),0); | |
244 | set_idt(phys_to_virt(0),0); | |
245 | ||
246 | /* now call it */ | |
3ab83521 HY |
247 | image->start = relocate_kernel_ptr((unsigned long)image->head, |
248 | (unsigned long)page_list, | |
249 | image->start, cpu_has_pae, | |
250 | image->preserve_context); | |
3122c331 HY |
251 | |
252 | #ifdef CONFIG_KEXEC_JUMP | |
253 | if (kexec_image->preserve_context) | |
254 | restore_processor_state(); | |
255 | #endif | |
256 | ||
257 | __ftrace_enabled_restore(save_ftrace_enabled); | |
5033cba0 | 258 | } |
1a3f239d | 259 | |
fd59d231 KO |
260 | void arch_crash_save_vmcoreinfo(void) |
261 | { | |
92df5c3e | 262 | #ifdef CONFIG_NUMA |
bcbba6c1 KO |
263 | VMCOREINFO_SYMBOL(node_data); |
264 | VMCOREINFO_LENGTH(node_data, MAX_NUMNODES); | |
fd59d231 KO |
265 | #endif |
266 | #ifdef CONFIG_X86_PAE | |
bcbba6c1 | 267 | VMCOREINFO_CONFIG(X86_PAE); |
fd59d231 KO |
268 | #endif |
269 | } | |
270 |